Vehicle control method and device based on interactive lane changing

Abstract

The application provides a vehicle control method and device based on interactive lane changing, and relates to the technical field of vehicle control. The method comprises the following steps: acquiring a predicted trajectory; constructing a root node of an initial search tree according to the current state of the ego vehicle and the current state of a target rear vehicle in the predicted trajectory; performing multiple rounds of iteration according to the root node in an acceleration interval constructed for the ego vehicle and the target rear vehicle to obtain a target search tree comprising multiple child nodes, wherein each round of iteration extends the child nodes downward, and the node value of the child nodes obtained by the extension and the node value of the nodes associated with the child nodes obtained by the extension are determined, and each child node comprises constructed state information of the ego vehicle and the rear vehicle; selecting a target child node from the multiple child nodes of the root node according to the node value of each child node; and controlling the ego vehicle to change lanes according to the constructed state information contained in the target child node, so that the ego vehicle is located in a target lane.

Description

Technical Field

[0001] This invention relates to the field of vehicle control technology, and in particular to a vehicle control method and apparatus based on interactive lane changing. Background Technology

[0002] The rapid development of autonomous driving technology has brought transformative opportunities to intelligent transportation systems. Lane changing is often one of the core driving behaviors performed by autonomous vehicles in complex traffic environments, especially in urban road scenarios. Vehicles often need to complete lane changing maneuvers under conditions of traffic congestion and heavy traffic flow. At this time, a dynamic game relationship is formed between the autonomous vehicle and the vehicle behind in the target lane. The autonomous vehicle attempts to merge into the target lane, while the vehicle behind may choose to slow down to give way or accelerate to refuse. The complexity and uncertainty of this two-way interaction make lane changing decisions extremely challenging. Existing technologies for lane changing decisions in autonomous driving typically use the Monte Carlo tree search method. However, in practice, the vehicle behind in the target lane is usually treated as a passive moving body. A two-way interaction model between the autonomous vehicle and the vehicle behind in the target lane is not established during the construction of the search tree. This fails to truly reflect the interactive characteristics of the vehicle behind responding to the autonomous vehicle's actions, resulting in poor smoothness of the lane changing trajectory. Summary of the Invention

[0003] This invention provides a vehicle control method and apparatus based on interactive lane changing, which solves the technical problem in the related art of poor smoothness of lane changing trajectory due to ignoring the actions of following vehicles during the lane changing planning process.

[0004] In a first aspect, embodiments of the present invention provide a vehicle control method based on interactive lane changing, the method comprising: Obtain the predicted trajectory; Based on the current state of the vehicle and the current state of the target vehicle behind it in the predicted trajectory, the root node of the initial search tree is constructed, where the target vehicle behind it is the vehicle currently located behind the vehicle in the target lane. Within the acceleration range constructed for the vehicle and the target vehicle, multiple iterations are performed based on the root node to obtain a target search tree including multiple child nodes. In each iteration, the child nodes are expanded downwards, and the node value of the expanded child node and the node value of the node associated with the expanded child node are determined. Each child node includes the construction status information of the vehicle and the vehicle. Based on the node value of each child node, the target child node is selected from the multiple child nodes of the root node; The vehicle is controlled to change lanes based on the construction status information contained in the target sub-node, so that the vehicle is located in the target lane.

[0005] Secondly, embodiments of the present invention provide a vehicle control device based on interactive lane changing, the device comprising: The acquisition module is used to acquire the predicted trajectory; A construction module is used to construct the root node of an initial search tree based on the current state of the self-vehicle and the current state of the target following vehicle in the predicted trajectory. The target following vehicle is the vehicle currently located behind the self-vehicle in the target lane. Within the acceleration interval constructed for the self-vehicle and the target following vehicle, multiple iterations are performed based on the root node to obtain a target search tree including multiple child nodes. In each iteration, the child nodes are expanded downwards, and the node value of the expanded child node and the node value of the node associated with the expanded child node are determined. Each child node includes the construction state information of the self-vehicle and the following vehicle. The selection module is used to select a target child node from multiple child nodes of the root node based on the node value of each child node. The control module is used to control the vehicle to change lanes based on the construction status information contained in the target sub-node, so that the vehicle is located in the target lane.

[0006] Thirdly, embodiments of the present invention provide a vehicle, including: processor; A memory for storing instructions to be executed by the processor, wherein the processor is configured to execute the instructions to implement the method as described in the first aspect.

[0007] Fourthly, embodiments of the present invention provide a storage medium that, when instructions in the storage medium are executed by a processor of an electronic device, enables the vehicle to perform the method as described in the first aspect.

[0008] Fifthly, embodiments of the present invention provide a computer program product, the program product comprising a computer program, the computer program being executed by a processor as described in the first aspect.

[0009] According to embodiments of the present invention, a vehicle control method based on interactive lane changing is provided, which involves: acquiring a predicted trajectory; constructing the root node of an initial search tree based on the current state of the vehicle and the current state of the target following vehicle in the predicted trajectory, wherein the target following vehicle is the vehicle currently located behind the vehicle in the target lane; performing multiple iterations within an acceleration interval constructed for the vehicle and the target following vehicle, based on the root node, to obtain a target search tree including multiple child nodes, wherein each iteration expands the child nodes downwards and determines the node value of the expanded child nodes and the node value of the nodes associated with the expanded child nodes, wherein each child node includes the construction state information of the vehicle and the following vehicle; selecting a target child node from the multiple child nodes of the root node based on the node value of each child node; and controlling the vehicle to change lanes based on the construction state information contained in the target child node, so that the vehicle is located in the target lane. During the iteration process based on the root node, different accelerations are constructed for both the vehicle and the target following vehicle. The child nodes are then expanded using combinations of these accelerations, so that each child node corresponds to a set of accelerations for both the vehicle and the target following vehicle. This allows the interaction between the vehicle and the following vehicle to be simultaneously reflected in the child nodes of the target search tree. Through multiple iterations, various interaction evolution paths under different combinations of acceleration choices for the vehicle and different response modes of the following vehicle are dynamically explored. This makes lane-changing decisions no longer dependent on fixed predictions of the following vehicle, but rather on the node value of each child node in the target search tree. The target child node is selected to control the vehicle to change lanes, thereby effectively improving the technical problem of poor smoothness of lane-changing trajectories caused by ignoring the response of the following vehicle during lane-changing planning. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 The flowchart of a vehicle control method based on interactive lane changing provided by some embodiments of the present invention is shown; Figure 2 The diagram illustrates a trajectory point sequence provided by some embodiments of the present invention; Figure 3 The structure of a vehicle control device based on interactive lane changing provided in some embodiments of the present invention is shown; Figure 4 The structure of a vehicle provided by some embodiments of the present invention is shown. Detailed Implementation

[0012] As described in the background section, the rapid development of autonomous driving technology has brought transformative opportunities to intelligent transportation systems. Lane changing is often one of the core driving behaviors performed by autonomous vehicles in complex traffic environments, especially in urban road scenarios. Vehicles often need to complete lane changing maneuvers under conditions of traffic congestion and heavy traffic flow. At this time, a dynamic game relationship is formed between the autonomous vehicle and the vehicle behind in the target lane. The autonomous vehicle attempts to merge into the target lane, while the vehicle behind may choose to slow down to give way or accelerate to refuse. The complexity and uncertainty of this two-way interaction make lane changing decisions extremely challenging. Existing technologies for lane changing decisions in autonomous driving typically use the Monte Carlo tree search method. However, in practice, the vehicle behind in the target lane is usually treated as a passive moving body. A two-way interaction model between the autonomous vehicle and the vehicle behind in the target lane is not established during the construction of the search tree. This fails to truly reflect the interactive characteristics of the vehicle behind responding to the autonomous vehicle's actions, resulting in poor smoothness of the lane changing trajectory.

[0013] According to embodiments of the present invention, a vehicle control method based on interactive lane changing is provided. During the iteration process based on the root node, different accelerations are constructed for the self-vehicle and the target following vehicle, and the constructed acceleration combinations are used to expand the child nodes, so that each child node corresponds to a set of self-vehicle acceleration and target following vehicle acceleration. Thus, the interaction relationship between the self-vehicle and the following vehicle can be reflected simultaneously in the child nodes of the target search tree. Through multiple rounds of iteration, multiple interactive evolution paths under different combinations of self-vehicle acceleration selection and different response modes of the following vehicle are dynamically explored. This makes the lane changing decision no longer dependent on fixed prediction of the following vehicle, but based on the node value of each child node in the target search tree, selecting the target child node to control the self-vehicle to change lanes. This effectively improves the technical problem of poor smoothness of lane changing trajectory caused by ignoring the action response of the following vehicle during the lane changing planning process in related technologies.

[0014] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0015] In this embodiment of the invention, the vehicle control method based on interactive lane changing provided by this embodiment can be executed by a target device. The target device can be a vehicle with an electronic control unit and intelligent computing capabilities. The vehicle may include a trajectory prediction module, an interactive search and planning module, and a motion planning module. The trajectory prediction module generates a predicted trajectory based on the current state of the vehicle and the current state of a vehicle behind it in the target lane. Then, it sends the predicted trajectory and lane-changing decision information indicating whether the vehicle should change lanes to the interactive search and planning module. After receiving the lane-changing decision information indicating whether the vehicle should change lanes, the interactive search and planning module plans according to the predicted trajectory sent by the trajectory prediction module, executes the vehicle control method based on interactive lane changing provided by this embodiment of the invention, and obtains a control strategy that causes the vehicle to change lanes to the target lane. The motion planning module then executes the control strategy to realize the vehicle lane change.

[0016] In an embodiment of the present invention, Figure 1 The flowchart of a vehicle control method based on interactive lane changing provided by some embodiments of the present invention is shown. For example... Figure 1 As shown, the vehicle control method based on interactive lane changing provided in this embodiment of the invention includes steps 110 to 150.

[0017] Step 110: Obtain the predicted trajectory.

[0018] In this embodiment of the invention, the predicted trajectory is generated by the vehicle's trajectory prediction module based on the vehicle's current motion state. It is a trajectory sequence used to characterize the vehicle's motion trend over a future period of time. The trajectory prediction module determines the vehicle's current position and speed based on the perception information collected by the sensors on the vehicle, and determines the target lane in combination with the vehicle's lane-changing target, thereby generating a predicted trajectory sequence for the vehicle during the lane-changing process. The target lane is the lane that the vehicle plans to switch into and is adjacent to the lane where the vehicle is currently located.

[0019] Step 120: Based on the current state of the vehicle and the current state of the target vehicle in the predicted trajectory, construct the root node of the initial search tree, where the target vehicle is the vehicle currently located behind the vehicle in the target lane.

[0020] In this embodiment of the invention, the predicted trajectory includes not only the current state of the vehicle itself but also the current state of the target following vehicle. The current state of the vehicle reflects its operational status, such as its position, speed, and heading angle. Similarly, the current state of the target following vehicle also reflects its operational status, such as its position and speed. The target following vehicle is the vehicle in the target lane that is closest to the vehicle in the longitudinal direction. Since the vehicle does not affect the behavior of the vehicle in front during the lane-changing game, and the responsibility of the vehicle in front is relatively low during the lane-changing process, this embodiment of the invention does not consider the vehicle in front during lane-changing control.

[0021] In this embodiment of the invention, the current state of the self-vehicle and the current state of the target vehicle in the predicted trajectory can be used as the root node of the initial search tree. The initial search tree can be used for subsequent Monte Carlo tree search. Each node represents the joint motion state of the self-vehicle and the target vehicle at a certain sampling time, including the position, velocity and acceleration information of the self-vehicle and the target vehicle, so as to reflect the interaction relationship between the self-vehicle and the target vehicle at the same time during the search process.

[0022] Step 130: Within the acceleration range constructed for the self-vehicle and the target following vehicle, perform multiple iterations based on the root node to obtain a target search tree including multiple child nodes. In each iteration, the child nodes are expanded downwards, and the node value of the expanded child node and the node value of the node associated with the expanded child node are determined. Each child node includes the construction status information of the self-vehicle and the following vehicle.

[0023] In this embodiment of the invention, the acceleration intervals constructed for the self-vehicle and the target following vehicle are used to limit the range of acceleration values ​​that can be assumed for the self-vehicle and the target following vehicle when expanding the sub-nodes in each iteration. The acceleration intervals can be preset according to the vehicle's physical braking capability, ride comfort, and typical driving behavior statistics. For example, the self-vehicle acceleration interval can be set to [-3m / s², 3m / s²], and the target following vehicle acceleration interval can be set to [-3m / s², 3m / s²]. Discrete sampling is performed at intervals of 0.5m / s² within each acceleration interval, thereby forming multiple acceleration combinations between the self-vehicle acceleration options and the target following vehicle acceleration options. Each acceleration combination corresponds to an interaction mode of self-vehicle intention and following vehicle response, so that the search tree can simultaneously cover multiple interactive evolution branches of different self-vehicle decisions and different following vehicle reactions during the expansion process.

[0024] In this embodiment of the invention, multi-round iteration refers to the iterative process of repeatedly expanding child nodes downwards from the root node or an expanded node and updating node values ​​based on the initial search tree. Each round of iteration expands child nodes by selecting a node in the current search tree as the parent node and generating a new set of child nodes for that parent node based on the acceleration interval mentioned above. Each newly generated child node corresponds to a combination of accelerations of the self-vehicle and the target following vehicle, and the joint motion state of the self-vehicle and the target following vehicle at the next sampling time is recursively determined. By expanding downwards layer by layer, the depth of the search tree gradually increases until it reaches the number of layers corresponding to the planned time domain. The node value is used to measure the quality of the joint state of the self-vehicle and the target following vehicle corresponding to the node during the lane-changing process, and is used to guide the node selection direction in subsequent iterations and the selection of the target child node in step 140. The construction of state information refers to the joint motion state of the self-vehicle and the target following vehicle stored in each child node at the corresponding sampling time of that node. This typically includes the self-vehicle's longitudinal position, lateral position, speed, acceleration, and heading angle, as well as the target following vehicle's longitudinal position, lateral position, speed, and acceleration. This ensures that each node in the search tree can independently reflect the interaction between the self-vehicle and the target following vehicle at that current node. Specifically, each tree node in the target search tree includes the following state information: .in These are the vehicle's lateral position, longitudinal position, vehicle speed, vehicle acceleration, and vehicle heading angle, respectively. These are the lateral position, longitudinal position, speed, and acceleration of the vehicle behind the target vehicle, respectively.

[0025] In this embodiment of the invention, starting from the root node, multiple acceleration combinations of the vehicle and the target vehicle are sampled within the acceleration range. The first layer of child nodes is generated recursively using a kinematic model, and the initial value of each child node is calculated. Subsequently, in multiple iterations, nodes are selected to continue expanding to deeper layers of child nodes. Each expansion layer is based on the construction state information of the parent node and the corresponding acceleration combination to recursively obtain the construction state information of the child node. The value of the child node is accumulated layer by layer to the parent node and nodes above it through backpropagation to update its statistical information until the preset number of iterations or the planned time domain depth is reached, and finally a target search tree that has fully explored multiple interactive evolution paths is obtained.

[0026] Step 140: Select the target child node from the multiple child nodes of the root node based on the node value of each child node.

[0027] In this embodiment of the invention, after multiple iterations, multiple first-layer child nodes have been extended below the root node. Each first-layer child node corresponds to a combination of accelerations of the vehicle and the target vehicle at the current sampling moment, representing a lane-changing strategy that the vehicle can choose at the current moment and the interaction mode with the target vehicle. The node value of each first-layer child node reflects the comprehensive cost evaluation obtained from multiple simulations based on this strategy. The higher the value, the better or more robust the strategy has been verified in multiple iterations. Since the root node represents the current motion state of the vehicle, the first-layer child nodes of the root node represent different lane-changing action options that the vehicle can directly execute at the current moment. Therefore, when controlling the vehicle to change lanes, one of the multiple first-layer child nodes of the root node can be selected as the target child node. The acceleration combination corresponding to the target child node is the lane-changing strategy that the vehicle should execute at the current moment.

[0028] Step 150: Control the vehicle to change lanes based on the construction status information contained in the target sub-node, so that the vehicle is located in the target lane.

[0029] In this embodiment of the invention, after determining the target sub-node, the vehicle construction status information stored in the target sub-node, including the vehicle's position, speed, acceleration, and other motion state parameters, can be used as the lane-changing control command for the current moment and output to the downstream motion planning module to control the vehicle to perform lane-changing actions according to the selected strategy.

[0030] In this embodiment of the invention, during the iteration process based on the root node, different accelerations are constructed for both the self-vehicle and the target following vehicle. The constructed acceleration combinations are used to expand the child nodes, so that each child node corresponds to a set of self-vehicle acceleration and target following vehicle acceleration. Thus, the interaction relationship between the self-vehicle and the following vehicle can be simultaneously reflected in the child nodes of the target search tree. Through multiple iterations, various interaction evolution paths under different combinations of self-vehicle acceleration selection and different response modes of the following vehicle are dynamically explored. This makes the lane-changing decision no longer dependent on a fixed prediction of the following vehicle, but based on the node value of each child node in the target search tree, selecting the target child node to control the self-vehicle to change lanes. This effectively improves the technical problem of poor smoothness of lane-changing trajectory caused by ignoring the action response of the following vehicle during the lane-changing planning process in related technologies.

[0031] In this embodiment of the invention, the interactive search part employs Monte Carlo tree search (MCTS). In the Monte Carlo search algorithm, each iteration mainly includes: selection, expansion, simulation, and backpropagation. Therefore, in step 130, during the multiple iterations based on the root node, each iteration includes the following steps: Step 131: Select the target extended node from the root node and the extended child nodes.

[0032] In this embodiment of the invention, the target extended node can be selected from the root node and the extended child nodes based on the node value, for example, the node with the highest node value can be selected as the target extended node; the target extended node can also be selected based on the cumulative number of visits, for example, in order to balance multiple interaction methods, the node with fewer cumulative visits can be selected as the target extended node.

[0033] Step 132: Within the acceleration range constructed for the vehicle and the target vehicle, extend multiple initial child nodes with the target extension node as the parent node.

[0034] In this embodiment of the invention, the expansion step is used to generate a new set of child nodes for the target expansion node, thereby broadening the search tree and exploring different decision branches. Specifically, discrete sampling is performed within the acceleration range of the self-vehicle and the acceleration range of the target following vehicle, and multiple acceleration sample values ​​of the self-vehicle are paired and combined with multiple acceleration sample values ​​of the target following vehicle to generate multiple sets of acceleration combinations. Each set of acceleration combinations corresponds to an interaction mode of self-vehicle intent and target following vehicle response, generating a corresponding initial child node.

[0035] Step 133: For each initial child node, simulate candidate paths in the planning time domain based on the assumed acceleration corresponding to the initial child node, and determine the total path value of the candidate paths.

[0036] In this embodiment of the invention, the purpose of the simulation phase is to start from the newly expanded initial sub-nodes, and to gradually deduce the motion state of the vehicle and the following vehicle at each sampling time according to the assumed acceleration corresponding to the initial sub-nodes within the planning time domain, thereby generating a complete candidate path and evaluating the total path value of the candidate path. The total path value represents the overall quality of the candidate path. The assumed acceleration corresponding to the initial sub-nodes characterizes the interaction mode between the vehicle and the target following vehicle at that node.

[0037] Step 134: Determine the node value of all nodes on the candidate path in reverse based on the total path value, and take the initial child node after updating the node value as the expanded child node. All nodes on the candidate path include the expanded child node and the node associated with the expanded child node.

[0038] In this embodiment of the invention, the purpose of the backpropagation phase is to determine the node values ​​of the expanded child nodes and the nodes associated with the expanded child nodes based on the total path value obtained. The nodes associated with the expanded child nodes are those nodes traversed layer by layer upwards along the candidate path. By backpropagating the total path value, the search direction in the selection phase can gradually converge towards higher-value nodes.

[0039] In this embodiment of the invention, the assumed acceleration corresponding to the initial child node includes a first assumed acceleration corresponding to the vehicle and a second assumed acceleration corresponding to the target following vehicle. In step 132, during the process of expanding multiple initial child nodes with the target expansion node as the parent node within the acceleration interval constructed for the vehicle and the following vehicle, one specific implementation includes: obtaining the first assumed acceleration located within the acceleration interval, wherein the first assumed acceleration is the acceleration required during the iteration process assuming the vehicle's intention is any one of acceleration lane-changing intention, constant speed lane-changing intention, and deceleration waiting intention; obtaining the second assumed acceleration located within the acceleration interval, wherein the second assumed acceleration is the acceleration required during the iteration process assuming the target following vehicle's response action is any one of proactive yielding action, following action, and acceleration rejection action; and pairing the first assumed acceleration with the second assumed acceleration to expand each of the initial child nodes with the target expansion node as the parent node.

[0040] In this embodiment of the invention, as exemplified above, during the process of expanding the sub-nodes for the acceleration intervals constructed for the self-vehicle and the target rear vehicle, the acceleration interval of the self-vehicle can be set to [-3m / s², 3m / s²], and the acceleration interval of the target rear vehicle can be set to [-3m / s², 3m / s²]. Discrete sampling is performed at intervals of 0.5m / s² within each acceleration interval. That is, the values ​​of the first assumed acceleration and the second assumed acceleration are only integers or have a decimal of 5.

[0041] In this embodiment of the invention, a first hypothetical acceleration located within the acceleration range is obtained. The first hypothetical acceleration is any acceleration within the range of [-3 m / s², 3 m / s²] that conforms to the discrete sampling standard. To make the first hypothetical acceleration more accurately represent the vehicle's intention, the vehicle's acceleration range can be further divided into a corresponding number of intervals based on the number of vehicle intentions. For example, in this embodiment of the invention, the vehicle's intention is any one of acceleration lane-changing intention, constant speed lane-changing intention, and deceleration waiting intention. Therefore, the vehicle's acceleration range can be divided into: an acceleration sub-interval [1 m / s², 3 m / s²] expressing the vehicle's intention to accelerate lane-changing, an acceleration sub-interval (-1 m / s², 1 m / s²) expressing the vehicle's intention to change lanes at a constant speed, and an acceleration sub-interval [-3 m / s², 1 m / s²] expressing the vehicle's intention to decelerate waiting. The corresponding vehicle intention is further determined based on the acceleration sub-interval into which the randomly determined first hypothetical acceleration falls. Accordingly, the acceleration range of the target vehicle can be divided into a corresponding number of acceleration sub-ranges based on the number of response actions of the target vehicle behind it. Then, the hypothetical response action of the target vehicle behind it is determined according to the sub-range into which the randomly determined second hypothetical acceleration falls. Subsequently, the intentions of the voluntary vehicle and the response actions of the target vehicle behind it are randomly matched to construct multiple interactive actions, thereby expanding the initial child node corresponding to each interactive action. For example, when the voluntary vehicle has 3 intentions and the target vehicle behind it has 3 response actions, 3*3 initial child nodes can usually be expanded when the target expansion node is the parent node.

[0042] In this embodiment of the invention, after selecting the target expansion node, the simulation step in the Monte Carlo tree search algorithm is entered. This usually refers to simulating based on a certain strategy until the search is completed, generating candidate paths and the total path value of the candidate paths.

[0043] In this embodiment of the invention, the interactive action corresponding to the initial child node can be used as a subsequent interactive strategy for simulation. A specific implementation of step 133, which involves simulating candidate paths within the planning time domain based on the assumed acceleration corresponding to the initial child node and determining the total path value of the candidate paths, includes: starting from the target expansion node, performing simulation processing at each sampling time in the planning time domain to obtain a single simulated child node, wherein the planning time domain includes N sampling times; and determining the candidate paths and the total path value of the candidate paths based on the construction status information of the root node, the target expansion node, and the multiple simulated child nodes obtained at multiple sampling times.

[0044] In this embodiment of the invention, starting from the target expansion node, the construction status information of the simulated child node at the next sampling time is recursively derived based on the construction status information of the target expansion node. The simulated child node is a virtual node temporarily generated during the simulation process, which does not participate in the node selection of subsequent iterations. It is only used in this simulation to represent the intermediate state of the candidate path at the corresponding sampling time. The construction status information corresponding to the simulated child node is also only used for this round of iteration and is automatically destroyed after the end of this round of iteration. A simulated child node is generated corresponding to each sampling time. The simulated child node does not expand to new nodes during the recursion process, but only deduces downwards along a single path. All simulated child nodes are connected in series along the time axis to form a complete candidate path from the root node, through the target expansion node to the planned time domain termination time. Since the target expansion node is located at a different level in the search tree, the number of remaining recursive sampling times is also different. Therefore, the number of simulated child nodes generated after the simulation is completed is determined by the level of the target expansion node. If the target expansion node is located at the kth level, then Nk sampling times need to be recursively generated from this node to the end time of the planning time domain, which will generate Nk simulated child nodes. These simulated child nodes are associated upwards along a single path during the recursion process, and finally connected into a complete candidate path. The total path value of the candidate path is calculated based on the construction status information included in the nodes on this complete candidate path.

[0045] In this embodiment of the invention, the simulation process includes the following steps: generating construction status information of simulated child nodes based on the construction status information of the parent node, wherein the target extended node is the parent node of the first simulation process; comparing whether the construction status information of the simulated child nodes meets the security conditions; if the construction status information of the simulated child nodes meets the security conditions, then updating the simulated child node to the parent node of the next simulation process, and using the simulated child node of the target extended node as the initial child node; if the construction status information of the simulated child nodes does not meet the security conditions, then canceling the simulated child node.

[0046] In this embodiment of the invention, in the first round of simulation processing, starting from the target expansion node, the target expansion node is used as the parent node for the first simulation processing. Based on the construction state information of the target expansion node and the first and second hypothetical accelerations included in the initial child node obtained in step 132, the construction state information of the initial child node is generated. At this time, the initial child node is also a simulated child node. It is determined whether the construction state information meets the safety conditions. If the safety conditions are met, the simulated child node is used as the initial child node that has been expanded into the search tree, and the construction state information of the initial child node can be persistently retained in the search tree for use in subsequent iterations. If the safety conditions are not met, the initial child node is canceled and the simulation of the current candidate path is terminated. In the second round of simulation processing, the initial child node is used as the parent node, and the simulation is recursively applied step by step within the planning time domain according to the time step. Each step of the recursion generates a simulated child node. The construction state information of the simulated child node is the joint motion state at the next sampling time obtained by recursively applying the construction state information of the parent node to the corresponding hypothetical acceleration. If the construction status information in the simulated child node meets the safety conditions, it is retained and used as the parent node for the next simulation process to continue the recursion downwards; if the construction status information in the simulated child node does not meet the safety conditions, the simulated child node is canceled and the recursion of the current candidate path is terminated, and simulated child nodes at subsequent sampling times on this path will no longer be generated.

[0047] In this embodiment of the invention, the construction state information of the parent node includes the first self-vehicle state information of the vehicle and the first rear-vehicle state information of the target vehicle, and the construction state information of the simulated child node includes the second self-vehicle state information of the vehicle and the second rear-vehicle state information of the target vehicle. During the recursive process in the simulation, the first self-vehicle state information in the simulated child node is typically determined by the first self-vehicle state information in the parent node, and correspondingly, the second rear-vehicle state information in the simulated child node is determined by the first rear-vehicle state information in the parent node. Specifically, generating the construction state information of the simulated child node based on the construction state information of the parent node includes: generating the second self-vehicle state information according to the first self-vehicle state information, the first assumed acceleration, and the trajectory point sequence of the vehicle, where the trajectory point sequence of the vehicle is obtained from the current state of the vehicle in the root node; and generating the second rear-vehicle state information based on the first rear-vehicle state information using an intelligent driver model. The first vehicle's status information includes its lateral position, longitudinal position, speed, acceleration, and heading angle. The first following vehicle's status information includes its lateral position, longitudinal position, speed, and acceleration. The second vehicle's status information includes its lateral position, longitudinal position, speed, acceleration, and heading angle. The second following vehicle's status information includes its lateral position, longitudinal position, speed, and acceleration. The trajectory point sequence includes the theoretical trajectory points of each vehicle at each time step within the planning time domain.

[0048] In an embodiment of the present invention, in the process of generating the second vehicle state information based on the first construction state information, the first assumed acceleration, and the lateral trajectory point sequence of the vehicle, a specific implementation includes: recursively obtaining the second vehicle longitudinal position using a vehicle kinematic model based on the first vehicle speed, the first vehicle longitudinal position, and the first assumed acceleration; filtering adjacent first and second trajectory points in the trajectory point sequence using the second vehicle longitudinal position; determining interpolation weights based on the relative position ratio of the second vehicle longitudinal position within the longitudinal interval formed by the first and second trajectory points; and performing linear interpolation on the first and second trajectory points based on the interpolation weights to obtain the second vehicle lateral position and the second vehicle heading angle.

[0049] In this embodiment of the invention, the vehicle kinematic model mainly includes the following formulas:

[0050]

[0051] in, The second vehicle speed, That is the speed of the first vehicle. Assuming acceleration as the first assumption, The time step between two adjacent sampling times. This is the second longitudinal position of the vehicle. This is the first longitudinal position of the vehicle.

[0052] Secondly, in the trajectory point sequence, adjacent first and second trajectory points are selected based on the second vehicle's longitudinal position. The trajectory point sequence is a discrete point sequence pre-generated based on a fifth-order polynomial, containing the theoretical lateral position, theoretical longitudinal position, and theoretical heading angle of the vehicle at each sampling time. The trajectory points are arranged monotonically in ascending order of longitudinal position. A first critical point in the lateral trajectory point sequence whose longitudinal position is not greater than and is closest to the second vehicle's longitudinal position, and a second critical point whose longitudinal position is not less than and is closest to the second vehicle's longitudinal position, are found. The first critical point is designated as the first trajectory point P1, and the second critical point is designated as the second trajectory point P2. Figure 2 The diagram illustrates the trajectory point sequences provided in some embodiments of the present invention, such as... Figure 2 As shown, the ego is the vehicle and the target rear is the vehicle behind it. The dashed arrows represent the trajectory point sequence, and P1 and P2 are two adjacent points on the trajectory point sequence determined by the longitudinal position of the first ego.

[0053] In this embodiment of the invention, when determining the interpolation weights based on the relative position ratio of the second vehicle's longitudinal position within the longitudinal interval formed by the first trajectory point and the second trajectory point, the interpolation weights can be determined using the bisection method with the following formula:

[0054] Where w is the interpolation weight. The theoretical longitudinal position of the first trajectory point. This represents the theoretical longitudinal position of the second trajectory point.

[0055] In this embodiment of the invention, during the process of linearly interpolating the first trajectory point and the second trajectory point according to the interpolation weights to obtain the second vehicle's lateral position and second vehicle's heading angle, the following formula can be used to determine the following:

[0056]

[0057] in, The second vehicle's lateral position. The theoretical lateral position of the first trajectory point. Let w be the theoretical lateral position of the second trajectory point, and w be the interpolation weight. The second vehicle heading angle, The theoretical heading angle of the first trajectory point. The theoretical heading angle of the second trajectory point.

[0058] In this embodiment of the invention, the current state of the vehicle in the root node includes the current vehicle speed, which determines the vehicle's current longitudinal speed and current lateral speed. Before generating the second vehicle state information based on the first vehicle state information, the first assumed acceleration, and the vehicle's trajectory point sequence, the method further includes a step of generating a trajectory point sequence. Specifically, a fifth-order polynomial is used to generate the theoretical lateral position of the vehicle at each sampling time in the planning time domain, resulting in N theoretical lateral positions. For each of the N theoretical lateral positions, based on the vehicle's current longitudinal speed, the corresponding theoretical longitudinal position and theoretical heading angle are generated, and the theoretical lateral position, the theoretical longitudinal position, and the theoretical heading angle are used as information of a theoretical trajectory point. Based on the N theoretical trajectory points, a trajectory point sequence is generated.

[0059] In this embodiment of the invention, time t can be used as a parameter to plan the time domain T (T=NΔt, where N is an integer and Δt is the time step between two adjacent sampling times). Using a fifth-order polynomial and preset boundary conditions, the theoretical lateral position of the vehicle at each sampling time within the planned time domain is described. :

[0060] Further based on the theoretical lateral position x at each sampling time... The theoretical longitudinal position corresponding to each sampling time is generated as follows:

[0061] in, This represents the theoretical longitudinal position of the vehicle at each sampling time. This represents the current longitudinal position of the vehicle in its current state. The current longitudinal speed is determined from the current speed in the current state of the vehicle.

[0062] The theoretical heading angle of the vehicle at each sampling time is determined by the trajectory tangent direction using the following formula:

[0063] in, This represents the theoretical heading angle of the vehicle at each sampling time. The lateral velocity of the vehicle at each sampling time is obtained by differentiating the fifth-order polynomial with respect to time. The current longitudinal speed is determined from the current speed in the current state of the vehicle.

[0064] Based on the above information, the trajectory point sequence is obtained:

[0065] In an embodiment of the present invention, in the process of generating the theoretical lateral position of the vehicle at each sampling time in the planning time domain using a fifth-order polynomial, a specific implementation includes: determining the coefficient values ​​of each term of the fifth-order polynomial according to preset boundary conditions, wherein the fifth-order polynomial is used to describe the curve of the lateral position of the vehicle changing with time; and generating the theoretical lateral position of the vehicle at each sampling time in the planning time domain based on the fifth-order polynomial after determining the coefficient values.

[0066] In this embodiment of the invention, the preset boundary conditions include start-point boundary conditions and end-point boundary conditions. When i=0, the current state of the vehicle in the root node is used as the theoretical lateral position of the vehicle in the start-point boundary conditions. Theoretical lateral velocity and theoretical lateral acceleration When i=N, it is assumed that the vehicle has reached the target lane; therefore, the lane width of the target lane is used as the theoretical lateral position in the endpoint boundary conditions. Considering the comfort of the vehicle's occupants, the theoretical lateral velocity in the endpoint boundary conditions is... =0, and theoretical acceleration It is 0.

[0067] In this embodiment of the invention, the coefficients of the fifth-degree polynomial are determined according to preset boundary conditions using the following formula:

[0068]

[0069]

[0070]

[0071]

[0072]

[0073] in .

[0074] In this embodiment of the invention, during the process of generating the second rear vehicle state information based on the first rear vehicle state information using the intelligent driver model, considering that the target rear vehicle remains unchanged when the driver cuts in, that is, the lateral position of the second rear vehicle is the same as the lateral position of the first rear vehicle, only the longitudinal update is performed on the rear vehicle. A specific longitudinal update implementation includes: when the target extended node is the parent node, determining the longitudinal position and speed of the second rear vehicle based on the second assumed acceleration, inputting the longitudinal position and speed of the second rear vehicle into the intelligent driver model to obtain a second corrected acceleration, determining the second rear vehicle acceleration included in the initial child node from the second assumed acceleration and the second corrected acceleration, and updating the second rear vehicle longitudinal position and speed in the initial child node based on the first rear vehicle longitudinal position, the first rear vehicle speed, and the second rear vehicle acceleration; when the initial child node or the simulated child node is the parent node, determining the second rear vehicle state information of the simulated child node of the parent node based on the first rear vehicle longitudinal position, the first rear vehicle speed, and the first rear vehicle acceleration.

[0075] In this embodiment of the invention, when the target extension node is the parent node, the second longitudinal position and the second speed of the rear vehicle are determined using the following formula based on the first rear vehicle longitudinal position, the first rear vehicle speed, and the second assumed acceleration:

[0076]

[0077] in, This represents the longitudinal position of the second rear vehicle in the initial child node. For the longitudinal position of the first rear vehicle in the target extended node, For the speed of the first vehicle following the target extended node, For the second assumed acceleration, For time step.

[0078] Furthermore, the longitudinal position and speed of the second following vehicle are input into the intelligent driver model to obtain the second corrected acceleration. The intelligent driver model (IDM) is a micro-car-following model used to simulate the longitudinal motion behavior of a vehicle on the road, emphasizing the balance between safety and efficiency.

[0079]

[0080] in, The second corrected acceleration is obtained through the intelligent driver model. The target is the maximum acceleration of the following vehicle. The desired speed of the following vehicle (usually the speed limit of the target lane). For the speed of the first vehicle following the target extended node, The simulated distance between the vehicle and the rear vehicle, calculated using the second assumed acceleration in the initial sub-node, is typically the longitudinal position of the second rear vehicle in the sub-node. Longitudinal position of the first rear vehicle in the parent node The difference between them; The safe following distance is obtained by using the speeds of the first and second vehicles in the target extension node; δ is the preset acceleration index (usually taken as 4).

[0081] In this embodiment of the invention, the second corrected acceleration represents the acceleration correction value of the following vehicle dynamically calculated by the intelligent driver model based on the speed of the first following vehicle in the target expansion node, the safe following distance, and the simulated distance between the self vehicle and the following vehicle in the initial sub-node. This value reflects the actual acceleration of the following vehicle in the current interaction state that better conforms to physical constraints. The second assumed acceleration represents a preset acceleration value sampled from the acceleration range, used as part of the action pair to characterize the macroscopic response intention of the following vehicle during the expansion phase. Therefore, after obtaining the second corrected acceleration, it is compared with the second assumed acceleration. If the deviation between the second corrected acceleration and the second assumed acceleration is greater than a threshold (typically 0.5 m / s²), the correction is applied. 2 If the second corrected acceleration is used as the second rear vehicle acceleration in the initial child node, and the deviation between the second corrected acceleration and the second assumed acceleration is not greater than the threshold, then the second assumed acceleration is kept as the second rear vehicle acceleration in the initial child node so that candidate paths can be derived from the second rear vehicle acceleration in the initial child node in subsequent simulations.

[0082] In this embodiment of the invention, if the second assumed acceleration is maintained as the acceleration of the second rear vehicle in the initial sub-node, then the longitudinal position and speed of the second rear vehicle in the initial sub-node are maintained. If the second corrected acceleration is used as the acceleration of the second rear vehicle in the initial sub-node, then the longitudinal position and speed of the second rear vehicle in the initial sub-node are corrected using the following formula:

[0083]

[0084] in, The longitudinal position of the second rear vehicle in the corrected initial child node. For the longitudinal position of the first rear vehicle in the target extended node, For the speed of the first vehicle following the target extended node, The corrected acceleration of the second rear vehicle. The speed of the second rear vehicle in the corrected initial child node.

[0085] In this embodiment of the invention, after comparison and correction, the longitudinal position, speed and acceleration of the second rear vehicle in the obtained initial child node are written into the construction status information of the initial child node in the search tree, so that the initial child node can be used as the parent node for subsequent recursion.

[0086] In this embodiment of the invention, when the initial child node or the simulated child node obtained from the previous recursion is used as the parent node in the second and subsequent recursions, the first rear vehicle acceleration in the initial child node (i.e., the second assumed acceleration that was just maintained) can be used. Or the corrected second rear vehicle acceleration The second rear vehicle acceleration continues as the simulated child node to simulate the interactive state in the initial child node, and the longitudinal position and speed of the second rear vehicle in the simulated child node of the simulated parent node are updated recursively using the following formula.

[0087]

[0088]

[0089] in, This represents the longitudinal position of the second rear vehicle in the simulated child nodes of the parent node. This represents the longitudinal position of the first rear vehicle in the parent node. The speed of the first vehicle following the parent node. The acceleration of the first vehicle following the parent node. The speed of the second vehicle in the simulated child node of the parent node.

[0090] In this embodiment of the invention, a security check must be performed on the simulated child node after each generation. One specific implementation method for comparing whether the construction status information of the simulated child node meets the security conditions includes: determining the simulated distance between the vehicle and the target vehicle based on the longitudinal position of the second vehicle and the longitudinal position of the second following vehicle; calculating a safe following distance based on the speed of the first vehicle and the speed of the first following vehicle, wherein the security conditions include the safe following distance; if the simulated distance is greater than the safe following distance, it indicates that the construction status information of the simulated child node meets the security conditions; if the simulated distance is not greater than the safe following distance, it indicates that the construction status information of the simulated child node does not meet the security conditions.

[0091] In this embodiment of the invention, the simulated sub-nodes include the difference between the longitudinal position of the second self-vehicle and the longitudinal position of the second rear vehicle, which is used as the simulated distance between the self-vehicle and the target rear vehicle. The safe following distance is calculated using the following formula:

[0092] in, To calculate the safe following distance, The preset basic safety distance is usually 2m. The speed of the first following vehicle. The preset safe workshop time interval, The relative speed between the two vehicles is the speed of the first vehicle and the speed of the first following vehicle. The preset maximum deceleration of the vehicle is typically taken as an absolute value between -4 and -6 m / s². The preset comfort deceleration is usually an integer value between -2 and -3 m / s².

[0093] In this embodiment of the invention, after generating a candidate path by deriving multiple simulated sub-nodes through multiple rounds of simulation, the total path value of the candidate path can be further generated based on the temporary construction state information included in each simulated sub-node. Specifically, step 134, determining the total path value of the candidate path, includes, in one specific implementation, calculating the comfort cost, security cost, and continuity cost of the candidate path respectively; and performing a weighted summation of the comfort cost, security cost, and continuity cost of the candidate path to obtain the total path value.

[0094] In this embodiment of the invention, the total path value of the candidate path can be calculated using the following formula:

[0095] in, The comfort weighting was determined through parameter adjustments and real-vehicle calibration. This is the trajectory continuity weight, usually a constant value; For safety, the weight is usually large to ensure that the cost of a collision is dominant. All of the above weights are derived from real vehicle calibration. For the price of comfort, For the sake of continuity, for the sake of security They are obtained using the following formulas:

[0096] |

[0097] in, As a trade-off for comfort, the smoothness of the candidate path is evaluated based on the vehicle's turning angle, acceleration, jerk, and deviation from the reference line within the candidate path. Specifically, The turning angle of the vehicle at each trajectory point on the candidate path reflects the magnitude of the steering operation; denoted as , representing the vehicle's acceleration at each trajectory point on the candidate path, reflecting the degree of acceleration or deceleration; jerk represents the vehicle's jerk at each trajectory point on the candidate path, i.e., the rate of change of acceleration. ; This represents the lateral deviation of the vehicle from the reference line, reflecting the vehicle's accuracy in following the predetermined driving path. The continuity cost is used to evaluate the smoothness of the connection between the current frame candidate path and the previous frame output trajectory. It is determined based on the difference between the current frame candidate path and the previous frame output trajectory. Specifically, For the trajectory of the current frame, The difference between the trajectory output from the previous frame and the trajectory is calculated after normalization. As a safety trade-off, the safety level of candidate paths is evaluated based on the distances between each trajectory point and obstacles within the candidate paths. Specifically, This is the preset minimum safe distance, usually 1.5m. Let be the distance between the i-th trajectory point and the obstacle.

[0098] In this embodiment of the invention, after calculating the total path cost, step 134 can be executed to backpropagate the total path cost to update the information of all nodes on the candidate path. Specifically, step 134, which involves backpropagating the total path cost to determine the node value of all nodes on the candidate path based on the total path cost, includes: propagating the total path cost back along the nodes traversed by the candidate path to update the number of visits and cumulative cost of all nodes on the candidate path; and generating the node value of each node on the candidate path based on the updated number of visits and cumulative cost.

[0099] In this embodiment of the invention, node value is used to measure the overall merits of the state corresponding to that node in lane-changing decisions. A larger value indicates a better node, meaning that the candidate path originating from that node performs better in terms of safety, comfort, and continuity. Conversely, the total path cost is smaller, indicating a better candidate path. Therefore, there is a negative correlation between node value and total path cost. During backpropagation, the number of visits to each node on the candidate path increases by one, and the cumulative cost of each node increases the total path cost of this simulation. Through multiple iterations, the number of visits and cumulative costs of each node accumulate, the average cost (the negative of the divisor of the cumulative cost and the number of visits) gradually converges, and the node value tends to stabilize, thus providing a reliable statistical basis for node selection in subsequent iterations. After the target search tree is constructed, starting from the root node, the child nodes with the highest node value can be selected layer by layer. By connecting these nodes sequentially, a complete coarse trajectory can be formed, and the target child node is finally output. The target child node is the first-level child node of the root node, i.e., the interaction action at the first sampling time.

[0100] In this embodiment of the invention, after each round of backpropagation, the node value affects the selection step of the next round of iteration. Specifically, step 131, which involves selecting the target expansion node from the root node and the expanded child nodes, includes: for each expanded child node, calculating the upper confidence bound value of the expanded child node based on the node value of the expanded child node; starting from the root node, selecting the expanded child node with the largest upper confidence bound value in the current layer layer by layer until a child node that is not fully expanded is reached; and taking the child node that is not fully expanded as the target expansion node.

[0101] In embodiments of the present invention, the upper confidence bound (UCB) of the expanded node can typically be determined using the following formula:

[0102] in, For node value, L is the number of times the node has been visited, while L is the total number of times the node's parent node has been visited, and C is an adjustable parameter.

[0103] In this embodiment of the invention, during each selection process, starting from the root node, the expanded child node with the largest upper confidence bound value in the current layer is selected layer by layer until an unexpanded child node is reached; the unexpanded child node is then taken as the target expansion node. For example, in the first iteration, the node with the largest UCB in the first layer is selected. This node is already fully expanded, and there are nodes in the second layer. Therefore, along this node, the node with the largest UCB value in the second layer is selected as the target expansion node, and so on downwards. If in the first iteration, the node with the largest UCB in the first layer is selected, but this node is not fully expanded and there are no nodes in the second layer, then this node is taken as the target expansion node.

[0104] In this embodiment of the invention, by incorporating both the vehicle's acceleration and the target following vehicle's acceleration into the extended dimension of the search tree, the search process simultaneously explores multiple interactive evolution paths under different combinations of the vehicle's lane-changing strategies and the following vehicle's different response modes. This allows lane-changing decisions to move beyond a simple one-way prediction of the target following vehicle's future trajectory. Instead, it dynamically evaluates the feasibility of lane-changing under different interaction modes through multiple rounds of iterative search. This maintains the stability and continuity of the plan even when the target following vehicle's speed fluctuates or its behavioral intentions change, significantly improving the success rate of lane-changing. Furthermore, by introducing comfort and continuity costs into the total path cost, passenger comfort and trajectory smoothness are considered while ensuring lane-changing safety. This avoids a decline in driving experience caused by frequent lane changes or sudden acceleration and deceleration, achieving comprehensive optimization of safety, comfort, and success rate.

[0105] Figure 3 The diagram shows a structural diagram of a vehicle control device based on interactive lane changing according to some embodiments of the present invention, such as... Figure 3 As shown, the vehicle control device 300 based on interactive lane changing provided in this embodiment of the invention includes: Module 310 is used to acquire the predicted trajectory; The construction module 320 is used to construct the root node of an initial search tree based on the current state of the self-vehicle and the current state of the target following vehicle in the predicted trajectory. The target following vehicle is the vehicle currently located behind the self-vehicle in the target lane. Within the acceleration interval constructed for the self-vehicle and the target following vehicle, multiple iterations are performed based on the root node to obtain a target search tree including multiple child nodes. In each iteration, the child nodes are expanded downwards, and the node value of the expanded child node and the node value of the node associated with the expanded child node are determined. Each child node includes the construction state information of the self-vehicle and the following vehicle. The selection module 330 is used to select a target child node from multiple child nodes of the root node based on the node value of each child node. The control module 340 is used to control the vehicle to change lanes based on the construction status information contained in the target sub-node, so that the vehicle is located in the target lane.

[0106] It should be noted that the embodiments of the vehicle control device based on interactive lane changing in this specification and the embodiments of the vehicle control method based on interactive lane changing in this specification are based on the same inventive concept. Therefore, the specific implementation of this embodiment can be referred to the corresponding embodiments of the vehicle control method based on interactive lane changing mentioned above, and the repeated parts will not be described again.

[0107] Figure 4 This is a structural block diagram of a vehicle provided in an embodiment of this application. Figure 4 As shown, the vehicle provided in this application embodiment includes a processor 410 and a memory 420, the memory being used to store instructions executable by the processor; wherein, the processor is configured to execute the instructions to implement the vehicle control method based on interactive lane changing provided in any of the above method embodiments.

[0108] In an exemplary embodiment, the vehicle may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the vehicle control method based on interactive lane changing provided in any of the above method embodiments.

[0109] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory including instructions, which can be executed by a processor of a device to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc. When the instructions in this non-transitory computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform the vehicle control method based on interactive lane changing provided in any of the above-described method embodiments.

[0110] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the vehicle control method based on interactive lane changing provided in any of the above-described method embodiments.

[0111] The above description does not provide detailed technical specifications regarding the structure of each layer. However, those skilled in the art should understand that layers and regions of desired shapes can be formed using various technical means. Furthermore, to form the same structure, those skilled in the art can also design methods that are not entirely identical to those described above. Additionally, although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be advantageously combined.

[0112] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0113] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A vehicle control method based on interactive lane changing, characterized in that, The method includes: Obtain the predicted trajectory; Based on the current state of the vehicle and the current state of the target vehicle behind it in the predicted trajectory, the root node of the initial search tree is constructed, where the target vehicle behind it is the vehicle currently located behind the vehicle in the target lane. Within the acceleration range constructed for the vehicle and the target vehicle, multiple iterations are performed based on the root node to obtain a target search tree including multiple child nodes. In each iteration, the child nodes are expanded downwards, and the node value of the expanded child node and the node value of the node associated with the expanded child node are determined. Each child node includes the construction status information of the vehicle and the vehicle. Based on the node value of each child node, the target child node is selected from the multiple child nodes of the root node; The vehicle is controlled to change lanes based on the construction status information contained in the target sub-node, so that the vehicle is located in the target lane.

2. The method as described in claim 1, wherein the step of performing multiple iterations based on the root node, wherein, Each iteration includes the following steps: Select the target expansion node from the root node and the expanded child nodes; Within the acceleration range constructed for the vehicle and the target vehicle, multiple initial child nodes are extended with the target extension node as the parent node; For each initial child node, candidate paths are simulated in the planning time domain based on the assumed acceleration corresponding to the initial child node, and the total path value of the candidate paths is determined. The node value of all nodes on the candidate path is determined in reverse based on the total value of the path, and the initial child node with the updated node value is taken as the expanded child node. All nodes on the candidate path include the expanded child nodes and the nodes associated with the expanded child nodes.

3. The method as described in claim 2, characterized in that, The assumed acceleration corresponding to the initial child node includes the first assumed acceleration corresponding to the vehicle and the second assumed acceleration corresponding to the target vehicle behind it. Within the acceleration range constructed for the vehicle and the following vehicle, multiple initial child nodes are extended, with the target extended node as the parent node, including: Obtain the first hypothetical acceleration located in the acceleration range. The first hypothetical acceleration is the acceleration required during the iteration process if the vehicle's intention is assumed to be any one of the following: acceleration lane change intention, constant speed lane change intention, and deceleration waiting intention. Obtain the second hypothetical acceleration located in the acceleration range. The second hypothetical acceleration is the acceleration required during the iteration process if the response action of the target vehicle behind it is assumed to be any one of the following actions: active yielding action, following action, and acceleration rejection action. The first hypothetical acceleration is paired with the second hypothetical acceleration to expand each of the initial child nodes with the target expansion node as the parent node.

4. The method as described in claim 3, characterized in that, The step of simulating candidate paths in the planning time domain based on the assumed acceleration corresponding to the initial child nodes and determining the total path value of the candidate paths includes: Starting from the target extended node, simulation processing is performed at each sampling time in the planning time domain to obtain a single simulation sub-node, wherein the planning time domain includes N sampling times; The candidate path and the total path value of the candidate path are determined based on the construction status information of the root node, the target expansion node, and the multiple simulated child nodes obtained at multiple sampling times. The simulation process includes the following steps: Based on the construction status information of the parent node, the construction status information of the simulated child node is generated, and the target extended node is the parent node of the first simulation process; Compare whether the construction status information of the simulated child nodes meets the security conditions; If the construction status information of the simulated child node meets the safety conditions, the simulated child node is updated to the parent node of the next simulation process, and the simulated child node of the target extended node is used as the initial child node. If the construction status information of the simulated child node does not meet the security conditions, then the simulated child node is cancelled.

5. The method as described in claim 4, characterized in that, The construction status information of the parent node includes the first self-vehicle status information of the vehicle and the first rear vehicle status information of the target vehicle. The construction status information of the simulated child node includes the second self-vehicle status information of the vehicle and the second rear vehicle status information of the target vehicle. The generation of simulated child node construction status information based on the parent node's construction status information includes: The second vehicle state information is generated based on the first vehicle state information, the first assumed acceleration, and the trajectory point sequence of the vehicle. The trajectory point sequence of the vehicle is obtained from the current state of the vehicle in the root node. The intelligent driver model generates the second rear vehicle status information based on the first rear vehicle status information.

6. The method as described in claim 5, characterized in that, The first vehicle status information includes the vehicle speed and the longitudinal position of the first vehicle; the second vehicle status information includes the longitudinal position, the lateral position, and the heading angle of the second vehicle; the trajectory point sequence includes the theoretical trajectory points of the vehicle at each time step within the planning time domain; The step of generating the second vehicle state information based on the first vehicle state information, the first assumed acceleration, and the lateral trajectory point sequence of the vehicle includes: Based on the first vehicle speed, the first vehicle longitudinal position, and the first assumed acceleration, the second vehicle longitudinal position is recursively obtained using the vehicle kinematics model. In the trajectory point sequence, adjacent first trajectory points and second trajectory points are selected by the second longitudinal position of the vehicle; The interpolation weights are determined based on the relative position ratio of the second vehicle's longitudinal position within the longitudinal interval formed by the first trajectory point and the second trajectory point. Based on the interpolation weights, linear interpolation is performed on the first trajectory point and the second trajectory point to obtain the second vehicle's lateral position and the second vehicle's heading angle.

7. The method as described in claim 6, characterized in that, The current state of the vehicle in the root node includes the current vehicle speed, and the current vehicle speed includes the current longitudinal vehicle speed; Before generating the second vehicle state information based on the first vehicle state information, the first assumed acceleration, and the trajectory point sequence of the vehicle, the method further includes: Using a fifth-order polynomial, the theoretical lateral positions of the vehicle at each sampling time within the planning time domain are generated, resulting in N theoretical lateral positions. For each of the N theoretical lateral positions, based on the current longitudinal speed of the vehicle, the theoretical longitudinal position and theoretical heading angle corresponding to the theoretical lateral position are generated, and the theoretical lateral position, the theoretical longitudinal position and the theoretical heading angle are used as information of a theoretical trajectory point. Generate a sequence of trajectory points based on N theoretical trajectory points.

8. The method as described in claim 7, characterized in that, The process employs a fifth-order polynomial to generate the theoretical lateral position of the vehicle at each sampling time within the planning time domain, including: Based on preset boundary conditions, the coefficient values ​​of each term of the fifth-degree polynomial are determined. The fifth-degree polynomial is used to describe the curve of the lateral position of the vehicle changing over time. Based on the fifth-order polynomial with determined coefficient values, the theoretical lateral position of the vehicle at each sampling time within the planning time domain is generated.

9. The method as described in claim 8, characterized in that, The first rear vehicle status information includes the longitudinal position of the first rear vehicle, the speed of the first rear vehicle, and the acceleration of the first rear vehicle; the second rear vehicle status information includes the longitudinal position of the second rear vehicle, the speed of the second rear vehicle, and the acceleration of the second rear vehicle. The step of generating the second rear vehicle status information based on the first rear vehicle status information using an intelligent driver model includes: When the target extended node is the parent node, the second rear vehicle longitudinal position and the second rear vehicle speed are determined according to the second assumed acceleration. The second rear vehicle longitudinal position and the second rear vehicle speed are input into the intelligent driver model to obtain the second corrected acceleration. The second rear vehicle acceleration included in the initial child node is determined from the second assumed acceleration and the second corrected acceleration. The second rear vehicle longitudinal position and the second rear vehicle speed in the initial child node are updated according to the first rear vehicle longitudinal position, the first rear vehicle speed and the second rear vehicle acceleration. When the initial child node or the simulated child node is the parent node, the second rear vehicle state information of the simulated child node of the parent node is determined based on the longitudinal position of the first rear vehicle, the speed of the first rear vehicle, and the acceleration of the first rear vehicle.

10. The method as described in claim 9, characterized in that, The comparison of whether the construction status information of the simulated child node meets the security conditions includes: Based on the longitudinal position of the second self-vehicle and the longitudinal position of the second rear vehicle, determine the simulated distance between the self-vehicle and the target rear vehicle; The safe following distance is calculated based on the speed of the first vehicle and the speed of the first following vehicle, and the safety conditions include the safe following distance. If the simulated distance is greater than the safe following distance, it indicates that the construction status information of the simulated sub-node meets the safety conditions; If the simulated distance is not greater than the safe following distance, it indicates that the construction status information of the simulated sub-node does not meet the safety conditions.

11. The method according to any one of claims 2-10, characterized in that, Determining the total path value of the candidate paths includes: Calculate the comfort cost, safety cost, and continuity cost of the candidate paths respectively; The total value of the path is obtained by weighted summing of the comfort cost, safety cost, and continuity cost of the candidate path.

12. The method according to any one of claims 2-10, characterized in that, The step of determining the node value of all nodes on the candidate path in reverse based on the total path value includes: The total value of the path is propagated backward along the nodes traversed by the candidate path, updating the number of visits and cumulative cost of all nodes on the candidate path; For each node on the candidate path, the node value is generated based on the updated number of visits and the cumulative cost of the node.

13. The method according to any one of claims 2-10, characterized in that, The step of selecting the target expansion node from the root node and the expanded child nodes includes: For each of the expanded child nodes, calculate the upper confidence bound value of the expanded child node based on the node value of the expanded child node; Starting from the root node, select the expanded child node with the largest upper confidence boundary value in the current layer layer by layer until a child node that has not been fully expanded is reached. The child nodes that are not fully expanded are taken as the target expansion nodes.

14. A vehicle control device based on interactive lane changing, characterized in that, The device includes: The acquisition module is used to acquire the predicted trajectory; A construction module is used to construct the root node of an initial search tree based on the current state of the self-vehicle and the current state of the target following vehicle in the predicted trajectory. The target following vehicle is the vehicle currently located behind the self-vehicle in the target lane. Within the acceleration interval constructed for the self-vehicle and the target following vehicle, multiple iterations are performed based on the root node to obtain a target search tree including multiple child nodes. In each iteration, the child nodes are expanded downwards, and the node value of the expanded child node and the node value of the node associated with the expanded child node are determined. Each child node includes the construction state information of the self-vehicle and the following vehicle. The selection module is used to select a target child node from multiple child nodes of the root node based on the node value of each child node. The control module is used to control the vehicle to change lanes based on the construction status information contained in the target sub-node, so that the vehicle is located in the target lane.

15. A vehicle, characterized in that, The vehicles include: processor; A memory for storing instructions to be executed by the processor, wherein the processor is configured to execute the instructions to implement the method as described in any one of claims 1 to 13.

Images