Vehicle automatic driving speed planning method, device and equipment and vehicle

Abstract

The invention relates to the technical field of automatic driving, and discloses a speed planning method, device and equipment for automatic driving of a vehicle and the vehicle, and the method comprises the steps: obtaining the driving information of the current vehicle in the automatic driving process of the current vehicle; determining motion information of the traffic participant based on the driving information; predicting track information of the traffic participants based on the motion information; generating a displacement time chart of the current vehicle based on the trajectory information, wherein the displacement time chart represents different time of the traffic participant and the displacement range of the current vehicle; determining a specified scene where the current vehicle is located, and determining a constraint range in the displacement time graph corresponding to the specified scene; inputting the constraint range into a pre-constructed action search model to obtain an optimal action sequence of the current vehicle; and inputting the optimal action sequence into a preset optimization algorithm, and planning to obtain a speed curve of the current vehicle.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular to a speed planning method, apparatus, device, and vehicle for autonomous driving. Background Technology

[0002] In the field of autonomous driving technology, vehicle speed planning is one of the core elements to ensure driving safety and traffic efficiency.

[0003] In existing technologies, many methods rely on rule-based decision-making systems or single optimization algorithm frameworks, which exhibit significant limitations when dealing with highly dynamic environments. Rule-based methods, due to their reliance on predefined logical conditions, struggle to handle unexpected situations not encoded in the rule set and lack the ability to generalize to unknown scenarios. Traditional optimization algorithms, such as gradient descent-based or numerical optimization methods, often suffer from high computational complexity when the solution space is high or non-convex, making it difficult to obtain the global optimum under real-time constraints. They may even get trapped in local optima, resulting in planning results that do not meet the safety requirements of dynamic environments. Summary of the Invention

[0004] In view of the above problems, this disclosure provides a speed planning method, apparatus, device, and vehicle for autonomous driving, to solve the technical problem that existing speed planning algorithms exhibit significant limitations when dealing with highly dynamic environments. The technical solution is as follows: A speed planning method for autonomous driving of vehicles, the method comprising: During the current autonomous driving process of the vehicle, the driving information of the current vehicle is acquired, including path information, first location information and first speed information; Based on the driving information, the motion information of the traffic participants is determined, including second position information, second speed information, and motion direction information; Based on the motion information, predict the trajectory information of the traffic participants; Based on the trajectory information, a displacement-time map of the current vehicle is generated, which represents the displacement range of the traffic participant relative to the current vehicle at different times. Determine the specified scenario in which the current vehicle is located, and determine the constraint range in the displacement-time graph corresponding to the specified scenario; The constraint range is input into a pre-built action search model to obtain the optimal action sequence for the current vehicle; The optimal action sequence is input into a pre-set optimization algorithm to plan the speed curve of the current vehicle.

[0005] It should be noted that this application first acquires motion information of the vehicle and traffic participants and predicts their trajectories to construct a displacement-time graph reflecting future interaction relationships, thereby transforming dynamic environmental information into a unified spatiotemporal representation framework. Then, by identifying specified scenarios and extracting corresponding constraint ranges, semantic scene rules are transformed into computable mathematical constraints, effectively guiding the subsequent decision-making process. Subsequently, the constraint ranges are input into an action search model, enabling efficient searching of feasible and safe action sequences within a constrained solution space, significantly reducing the computational complexity and local optima risk associated with traditional optimization methods due to excessively large search spaces. Finally, optimization algorithms smooth and refine the action sequences, generating a safe and reliable speed curve that conforms to vehicle dynamics. This organic combination of technical steps not only improves the real-time performance and reliability of speed planning in dynamic environments but also enhances the system's adaptability to complex interaction scenarios and the safety of its decisions.

[0006] Optionally, determining the specified scenario in which the current vehicle is located includes: Obtain traffic sign information at a preset distance ahead of the current vehicle; The designated scenario is determined based on the driving information, the motion information, and the traffic sign information.

[0007] It should be noted that this application obtains traffic sign information within a preset distance ahead of the vehicle and integrates it with the vehicle's driving status and the movement information of surrounding participants to accurately identify the specific scenario (such as intersections, speed-limited zones, pedestrian crossings, etc.) in which the vehicle is currently located. This mechanism allows the system to move beyond relying on single geometric or kinematic information and instead incorporate high-level semantic rules, thereby transforming abstract traffic regulations and road constraints into calculable and operable scenario types. Based on this specified scenario, the system can accurately invoke the corresponding displacement-time graph constraint range, ensuring that subsequent action search and speed optimization strictly adhere to traffic rules and safety criteria. This effectively avoids decision-making biases caused by scenario misjudgment or rule omissions, ultimately enhancing the compliance, environmental adaptability, and overall system reliability of the speed planning results.

[0008] Optionally, the step of inputting the constraint range into a pre-built action search model to obtain the optimal action sequence of the current vehicle includes: Within a pre-defined decision time range, the longitudinal motion behavior of the current vehicle is discretized into multiple acceleration actions; Multiple acceleration actions are input into an action search model constructed by a Monte Carlo tree search algorithm, and search optimization is performed within the constraints to obtain the optimal acceleration action sequence that satisfies all constraints.

[0009] It should be noted that this application significantly improves the decision-making quality and efficiency of velocity planning in complex dynamic environments by discretizing longitudinal motion behavior into multiple acceleration actions and using the Monte Carlo Tree Search (MCTS) algorithm to search within a predefined constraint range. This method transforms the continuous decision space into a finite, manageable set of actions through discretization, providing a structured search foundation for MCTS and effectively overcoming the high computational complexity and susceptibility to local optima inherent in traditional optimization algorithms in high-dimensional continuous spaces. Simultaneously, by directly embedding the constraint range extracted from the displacement-time graph into the MCTS search process, it ensures that the algorithm explores only within the feasible solution space that meets safety and rule requirements, greatly reducing invalid search paths and enhancing the targeting and safety of the decision. Ultimately, MCTS can output a globally superior acceleration action sequence within a preset decision time, thereby providing a high-quality initial solution for subsequent optimization algorithms and comprehensively improving the reliability, real-time performance, and adaptability to dynamic interactive scenarios of the velocity planning results.

[0010] Optionally, the construction of the action search model includes: Determine the state set, action set, and reward function of the action search model; wherein, The state set includes the current speed and displacement of the vehicle; The set of actions includes rapid acceleration, rapid deceleration, constant speed, gradual acceleration, and gradual deceleration; The reward function is constructed based on the minimum distance to the traffic participant and the first speed information, and is used to evaluate the quality of the action sequence. The higher the reward value, the better the corresponding action sequence.

[0011] It should be noted that this application significantly improves the scientific rigor, safety, and efficiency of the speed planning decision-making process by constructing an action search model with clearly defined states, a reasonable action space, and a targeted reward function. This model abstracts the vehicle state into a set containing speed and displacement, comprehensively representing the vehicle's real-time motion characteristics. The action set covers typical longitudinal driving behaviors such as rapid acceleration, rapid deceleration, constant speed, gradual acceleration, and gradual deceleration, ensuring the coverage of the decision-making process while reducing search complexity through discretization. Crucially, the reward function is constructed based on the minimum distance to traffic participants and the vehicle's speed, effectively transforming the requirements for maintaining safe distances and smooth driving into quantifiable optimization objectives. This guides the Monte Carlo tree search algorithm to quickly select acceleration action sequences that ensure safe distances while also considering driving efficiency and comfort within constraints. This mechanism not only avoids the blindness of unconstrained random searches but also strengthens the targeted response to key risk factors in dynamic interaction scenarios, ultimately significantly enhancing the overall quality, reliability, and environmental adaptability of the speed planning results.

[0012] Optionally, the step of inputting the constraint range into a pre-built action search model to obtain the optimal action sequence of the current vehicle includes: When the Monte Carlo tree search algorithm expands the search tree, a branch reduction operation is performed to generate a sequence of candidate actions that meet the constraints. The constraints include: the first speed does not exceed a preset speed limit, there is no collision with the traffic participant, and the distance to the forward traffic participant meets the safe distance. The candidate action sequences are randomly simulated to predict the motion state of the current vehicle under different candidate action sequences, and the optimal action sequence of the current vehicle is determined based on the motion state.

[0013] It should be noted that this application significantly improves the efficiency, safety, and decision reliability of action sequence search by deeply integrating constraints into the expansion and simulation stages of the Monte Carlo Tree Search algorithm. This method actively performs a branch reduction operation when expanding the search tree, strictly filtering action branches based on preset key constraints such as speed limits, collision avoidance, and safe distances. This fundamentally eliminates a large number of invalid actions that violate traffic rules or pose safety hazards, greatly compressing the search space and avoiding the waste of computational resources and decreased timeliness caused by blindly exploring all possible actions in traditional MCTS. Subsequently, by randomly simulating the candidate action sequences generated after branch reduction, the future motion state of the vehicle under different action sequences is accurately predicted. This allows for the evaluation of sequence quality within the feasible solution space based on actual motion effects rather than simply heuristic rules. This ensures that the final optimal action sequence not only theoretically satisfies all hard constraints but also exhibits good dynamic adaptability and practical feasibility in the simulation environment, thereby comprehensively enhancing the real-time performance, safety, and overall decision quality of the speed planning results.

[0014] Optionally, the random simulation of the candidate action sequence includes at least one of the following modes: A non-interactive trajectory mode is adopted, in which the trajectories of the traffic participants based on the prediction are reflected as static constraints in the displacement-time graph; Alternatively, an interactive mode based on a vehicle-following model can be adopted, in which the state of the traffic participant is updated in real time according to the current vehicle's action at each extended node of the Monte Carlo tree search; wherein the traffic participant calculates its future motion state based on the behavior of the current vehicle using the vehicle-following model. Alternatively, a multi-vehicle cooperative mode can be adopted, in which the current vehicle plays the leading role and the traffic participants play the following role, and the optimal response strategy of the traffic participants is dynamically calculated based on the actions of the current vehicle.

[0015] It should be noted that this application significantly improves the decision-making intelligence, safety, and adaptability of autonomous driving speed planning in complex interactive scenarios by introducing a multi-mode stochastic simulation mechanism. This method provides three simulation modes: non-interactive trajectory, interactive mode based on a following model, and multi-vehicle cooperative mode. These modes can be flexibly selected or combined according to actual scenario requirements, thus systematically solving the problems of single environmental modeling and lagging interactive feedback in traditional methods. The non-interactive mode treats traffic participants as static constraints, providing a basic safety boundary guarantee for decision-making. The interactive mode updates the status of surrounding vehicles in real time through a following model, enabling the simulation process to dynamically reflect the impact of the vehicle's behavior on other vehicles, enhancing the accuracy of trajectory prediction and the coupling of decision-making. The multi-vehicle cooperative mode further treats the vehicle as the dominant player, simulating the optimal response strategy of traffic participants, achieving efficient modeling of cooperative driving scenarios. This multi-level, configurable simulation framework allows the Monte Carlo tree search algorithm to evaluate the long-term effects of action sequences in a more realistic and dynamic environment, thereby generating optimal action sequences that meet immediate safety constraints while possessing good interactivity and foresight, comprehensively improving the reliability, robustness, and scenario generalization ability of the speed planning system.

[0016] Optionally, the step of inputting the optimal action sequence into a pre-set optimization algorithm to plan the speed curve of the current vehicle includes: Based on the optimal action sequence and the constraint range in the displacement-time graph, the optimization algorithm is used to solve for the smooth velocity curve. The cost function of the optimization algorithm takes into account the deviation from the path information and the cumulative value of acceleration. The constraints of the optimization algorithm include initial motion state, final motion state, displacement constraints, and kinematic constraints. The displacement constraints are derived from the upper and lower limits of the displacement corresponding to the optimal action sequence.

[0017] It should be noted that this application combines the optimal action sequence output by Monte Carlo tree search with the optimization algorithm, and introduces a cost function with clear engineering significance and multi-dimensional constraints, achieving a high-quality transformation from discrete action sequences to continuous and smooth velocity curves, significantly improving the comfort, accuracy, and feasibility of velocity planning. This method uses the upper and lower limits of displacement provided by the optimal action sequence as core displacement constraints, ensuring that the optimization process always remains within a safe and feasible solution space. Simultaneously, the cost function of the optimization algorithm considers both the tracking deviation on the predetermined path and the cumulative acceleration value, effectively balancing the two key performance indicators of driving accuracy and ride comfort. Furthermore, the strict adherence to the initial and final motion states and vehicle kinematic constraints ensures that the final generated velocity curve not only fully inherits the safety and rationality of the front-end decision-making module, but also achieves a deep integration of trajectory smoothness, power adaptability, and control precision in the continuous domain, thereby comprehensively enhancing the overall reliability, ride quality, and adaptability to complex dynamic environments when the autonomous driving system executes velocity planning.

[0018] A speed planning device for autonomous driving of vehicles, comprising: The information acquisition unit acquires the driving information of the current vehicle during the current autonomous driving process. The driving information includes path information, first location information, and first speed information. The information determination unit determines the motion information of the traffic participant based on the driving information, the motion information including second position information, second speed information and motion direction information; The prediction unit predicts the trajectory information of the traffic participant based on the motion information; The generation unit generates a displacement-time map of the current vehicle based on the trajectory information. The displacement-time map represents the displacement range of the traffic participant relative to the current vehicle at different times. The range determination unit determines the specified scenario in which the current vehicle is located and determines the constraint range in the displacement-time graph corresponding to the specified scenario; The sequence determination unit inputs the constraint range into a pre-built action search model to obtain the optimal action sequence of the current vehicle; The speed planning unit inputs the optimal action sequence into a pre-set optimization algorithm to plan the speed curve of the current vehicle.

[0019] A speed planning device for autonomous driving of vehicles, comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: During the current autonomous driving process of the vehicle, the driving information of the current vehicle is acquired, including path information, first location information and first speed information; Based on the driving information, the motion information of the traffic participants is determined, including second position information, second speed information, and motion direction information; Based on the motion information, predict the trajectory information of the traffic participants; Based on the trajectory information, a displacement-time map of the current vehicle is generated, which represents the displacement range of the traffic participant relative to the current vehicle at different times. Determine the specified scenario in which the current vehicle is located, and determine the constraint range in the displacement-time graph corresponding to the specified scenario; The constraint range is input into a pre-built action search model to obtain the optimal action sequence for the current vehicle; The optimal action sequence is input into a pre-set optimization algorithm to plan the speed curve of the current vehicle.

[0020] A vehicle, comprising the speed planning method for autonomous driving described above.

[0021] The above description is merely an overview of the technical solution disclosed herein. In order to better understand the technical means of this disclosure and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this disclosure more apparent and understandable, specific embodiments of this disclosure are described below. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating a speed planning method for autonomous driving of a vehicle provided in an embodiment of this disclosure is shown. Figure 2 This diagram illustrates the scenario decision output provided in an embodiment of the present disclosure. Figure 3 An interactive schematic diagram provided by an embodiment of this disclosure is shown; Figure 4A schematic diagram of the structure of a speed planning device for autonomous driving of a vehicle provided in an embodiment of this disclosure is shown; Figure 5 This diagram illustrates the structure of a speed planning device for autonomous driving of a vehicle, according to an embodiment of the present disclosure. Figure 6 A schematic diagram of the structure of a vehicle provided in an embodiment of this disclosure is shown. Detailed Implementation

[0023] In autonomous driving technology, vehicle speed planning is a crucial step in ensuring safe and efficient driving. However, existing autonomous driving speed planning methods suffer from several problems. On the one hand, some methods rely solely on single rule-based decisions or optimization algorithms, making it difficult to cope with complex and ever-changing traffic scenarios. For example, rule-based methods are prone to decision-making errors when faced with sudden changes in traffic participant behavior due to incomplete rule coverage; while optimization algorithm-based methods may fail to find the optimal solution within a limited time due to an excessively large search space, especially in scenarios with dynamic obstacle interactions, making it difficult to generate reasonable speed plans in real time.

[0024] On the other hand, existing technologies have shortcomings in balancing scenario constraints and decision-making randomness. While stochastic decision-making methods such as Monte Carlo Tree Search (MCTS) can explore multiple possibilities to some extent, in certain scenarios, such as when stopping at traffic lights or speed limits before a left turn, the lack of effective scenario constraints may lead to unexpected decision results, affecting driving safety. Furthermore, existing technologies do not adequately consider the interactivity of traffic participants. Most methods rely on fixed obstacle trajectory predictions, failing to consider the real-time impact of the vehicle's behavior on surrounding traffic participants. This results in speed planning being out of sync with the actual traffic environment, easily leading to collision risks.

[0025] Terminology Explanation: Monte Carlo Tree Search (MCTS): A heuristic search algorithm based on random sampling that explores possible sequences of actions by building a search tree, suitable for complex decision problems.

[0026] Vertical state machine: used to determine the traffic scenario in which the vehicle is located and generate corresponding decision constraints based on preset rules, such as speed limit and parking.

[0027] Time-Displacement Diagram (ST Diagram): A graph with time on the horizontal axis and displacement on the vertical axis, used to visually represent the movement trajectory and constraint relationships of vehicles and obstacles.

[0028] QP / IPOPT: Both are optimization algorithms, where QP is a quadratic programming algorithm and IPOPT is an interior point optimization algorithm, used to solve smooth velocity curves.

[0029] In summary, this technical solution aims to address the following problems existing in current autonomous driving speed planning methods: It struggles to cope with complex and ever-changing traffic scenarios, and single rule-based decision-making or optimization algorithms perform poorly in handling unexpected situations. The balance between scenario constraints and decision-making randomness is insufficient, potentially leading to unexpected decision outcomes in specific scenarios and impacting driving safety. Insufficient consideration of the interactivity of traffic participants and reliance on fixed obstacle trajectory predictions result in speed planning being out of touch with the actual traffic environment, posing a collision risk. Achieving both smoothness and optimality in speed planning is difficult; either excessive time consumption due to an overemphasis on finding the optimal solution, or an unsmooth speed curve due to simplified calculations, becomes problematic.

[0030] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0031] Therefore, this application provides a flowchart illustrating a speed planning method for autonomous driving of vehicles, as shown below. Figure 1 As shown, this process can be executed by a speed planning system. Certain input parameters or intermediate results within the process can be manually adjusted to help improve accuracy.

[0032] The method flow steps of this application embodiment are as follows: S101, during the current autonomous driving process of the vehicle, the driving information of the current vehicle is obtained, including path information, first location information and first speed information.

[0033] In the embodiments described in this specification, during the current autonomous driving process of the vehicle, the vehicle's driving information is acquired in real time through the onboard sensor system, high-precision positioning module, and preset path planner. Specifically, this information may be global path information provided by the path planner, first position information obtained by the fusion calculation of the Global Navigation Satellite System (GNSS) and Inertial Measurement Unit (IMU), and first speed information obtained by the onboard bus or wheel speed sensors.

[0034] S102, based on the driving information, determine the motion information of the traffic participant, the motion information including second position information, second speed information and motion direction information.

[0035] In the embodiments of this specification, based on the current vehicle driving information acquired in S101, the vehicle-mounted perception system (including cameras, lidar, millimeter-wave radar, etc.) is used to detect and track the surrounding environment, identifying traffic participants (such as surrounding vehicles, pedestrians, etc.) related to the current vehicle's driving. Then, through multi-sensor fusion technology, the motion information of each traffic participant is determined, including second position information calculated from their relative positions, second velocity information obtained through continuous frame target tracking, and motion direction information estimated based on historical motion data.

[0036] S103, predict the trajectory information of the traffic participant based on the motion information.

[0037] In the embodiments of this specification, based on the motion information of each traffic participant obtained in S102, a trajectory prediction algorithm (such as a physical model-based, machine learning model-based, or hybrid method) is used to predict their future behavior, generating trajectory information of the traffic participants over a period of time. This trajectory information describes the predicted position and motion state of the target at various future moments.

[0038] S104, Based on the trajectory information, generate a displacement-time map of the current vehicle, the displacement-time map representing the displacement range of the traffic participant relative to the current vehicle at different times.

[0039] In the embodiments of this specification, based on the trajectory information of traffic participants predicted in S103 and combined with the current motion state of the vehicle, a displacement-time diagram with the vehicle as the reference frame is constructed in a displacement-time coordinate system. This diagram transforms the spatiotemporal occupancy of traffic participants into a constrained region relative to the vehicle's displacement, intuitively representing the safe displacement range between traffic participants and the vehicle at different time points.

[0040] S105, determine the specified scenario where the current vehicle is located, and determine the constraint range in the displacement-time graph corresponding to the specified scenario.

[0041] In the embodiments of this specification, traffic sign information (such as traffic lights, signs, road markings, etc.) within a preset distance in front of the current vehicle is obtained through an onboard environmental perception system. This information is then combined with the vehicle's driving information and the movement information of traffic participants. A scene recognition algorithm (such as a rule-based or machine learning classifier) ​​is used to determine the specified scene (such as an intersection, ramp, pedestrian crossing, etc.) in which the current vehicle is located. Subsequently, based on the identified specified scene type, the corresponding constraint range is extracted and determined from the displacement-time map generated in S104. This range reflects the safety and regulatory requirements under the specific scene rules.

[0042] S106, Input the constraint range into the pre-built action search model to obtain the optimal action sequence of the current vehicle.

[0043] In the embodiments of this specification, the constraint range determined in S105 is input into a pre-built action search model (such as a decision model based on Monte Carlo tree search). Under the constraints of the constraint range, the model searches and evaluates the longitudinal action space of the current vehicle (such as acceleration, deceleration, maintaining speed, etc.), and finally outputs an action sequence that satisfies all constraints and is optimal according to the reward function.

[0044] S107, The optimal action sequence is input into a pre-set optimization algorithm to plan the speed curve of the current vehicle.

[0045] In the embodiments of this specification, the optimal action sequence obtained in S106 is used as input and fed into a pre-set optimization algorithm. This optimization algorithm is based on the macro-level guidance of action sequence planning, while comprehensively considering multiple objectives and constraints such as path tracking accuracy, ride comfort, and vehicle dynamics, to perform refined calculations and ultimately generate a smooth, feasible, and safe optimal speed curve.

[0046] It should be noted that this application first acquires motion information of the vehicle and traffic participants and predicts their trajectories to construct a displacement-time graph reflecting future interaction relationships, thereby transforming dynamic environmental information into a unified spatiotemporal representation framework. Then, by identifying specified scenarios and extracting corresponding constraint ranges, semantic scene rules are transformed into computable mathematical constraints, effectively guiding the subsequent decision-making process. Subsequently, the constraint ranges are input into an action search model, enabling efficient searching of feasible and safe action sequences within a constrained solution space, significantly reducing the computational complexity and local optima risk associated with traditional optimization methods due to excessively large search spaces. Finally, optimization algorithms smooth and refine the action sequences, generating a safe and reliable speed curve that conforms to vehicle dynamics. This organic combination of technical steps not only improves the real-time performance and reliability of speed planning in dynamic environments but also enhances the system's adaptability to complex interaction scenarios and the safety of its decisions.

[0047] Regarding the above, the overall algorithm architecture of this invention consists of four key steps: relevant traffic participant search, longitudinal state machine / decision, coarse speed planning based on MCTS, and speed planning based on optimization algorithms. Through the coordinated work of these four steps, a complete process from traffic environment perception to final speed curve generation is achieved.

[0048] The purpose of the relevant traffic participant search step is to locate traffic participants appearing on the main vehicle's path and incorporate them into the ST (Speed ​​Targeting) graph to provide foundational data for subsequent decision-making and planning. Specifically, sensors (such as radar and cameras) are used to perceive the surrounding traffic environment in real time, identify vehicles, pedestrians, traffic lights, and other traffic participants on the main vehicle's path, obtain information such as their position, speed, and direction of movement, predict their future trajectories, and map this information onto the ST graph to clarify the displacement range of each traffic participant at different time points, providing obstacle information for subsequent scenario decisions and speed planning.

[0049] The longitudinal state machine / decision is used to determine the scenario in which the vehicle is located and to generate corresponding constraints based on rules to limit the search range of MCTS and ensure that the expected behavior is generated under the expected scenario.

[0050] For scenario assessment and rule-based decision-making, the system determines the scenario the vehicle is in based on its current position, speed, the status of surrounding traffic participants, and traffic signs (such as traffic lights and stop signs). Examples include proceeding straight through an intersection, turning left, or encountering a stop sign. Within each scenario, decisions are made based on preset rules, generating corresponding constraints. For instance, the vehicle speed is limited before turning left; a stop sign requires waiting; a red traffic light necessitates waiting, while a green light allows passage.

[0051] To avoid frequent inter-frame decision fluctuations, a rule-based hysteresis mechanism can be adopted. Taking traffic light passage decisions as an example, the conditions for entering the stop-and-wait state are slightly more lenient, that is, entering the stop-and-wait state when driving at the current constant speed would result in running a red light; the conditions for exiting the stop-and-wait state are slightly more stringent, that is, exiting the stop-and-wait state only when driving at the minimum value (intersection speed limit, current speed) at a constant speed would allow passage.

[0052] The results of scenario decisions need to be transformed into constraints in the ST diagram, which mainly include the following forms: During certain time periods, in the ST diagram, the main vehicle must be below a certain obstacle (which may be virtual, such as the stop line position corresponding to a traffic light) (that is, the main vehicle is behind a certain obstacle), which is converted into the upper and lower limits of displacement (s) in the ST diagram.

[0053] The parking decision that the car should remain in a certain position after a certain moment (i.e., the final state).

[0054] Speed ​​limit sequence in certain scenarios, that is, the maximum value of speed (v) at each displacement (s). This value needs to take into account both the road speed limit and the speed limit in the scenario (such as deceleration before the intersection, and a lower speed limit value on the path from 30m away from the intersection to leaving the intersection).

[0055] For special handling of constraints, if the decision result is between t=4 and 5 seconds, the upper limit of s is 100; then, even though there is no decision during the time interval between t=3 and 4 seconds, the upper limit of s must still be set to 100, because the vehicle cannot reverse, to ensure the continuity and rationality of displacement. Similarly, if the decision result is between t=3 and 4 seconds, the lower limit of s is 20; then, even though there is no decision during the time interval between t=5 and 6 seconds, the lower limit of s must still be set to 20.

[0056] See the schematic diagram of the scenario decision output. Figure 2 The thick lines around obstacles B, C, and D represent the upper and lower limits of scenario 's'. This means the decision-making process can consider as many scenarios as possible, but doesn't need to consider every single obstacle or virtual obstacle. The decision of which path the main vehicle should take—above or below obstacle A (i.e., whether to get ahead of obstacle A or let it pass behind A)—is determined by MCTS. Since B, C, and D all participate in scenario decision-making, the upper and lower limits are based on them. A does not participate in scenario decision-making, so the MCTS doesn't know whether to go above or below A beforehand.

[0057] For coarse speed planning based on MCTS, within the decision time (5-8 seconds), vehicle behavior is discretized into a finite number of actions (accelerations). MCTS is used to find the optimal action sequence that satisfies the existing constraints, so as to realize functions such as passing / stopping decisions for avoiding vehicles / pedestrians / traffic lights.

[0058] First, we define the Markov decision process as follows: The state set includes the vehicle's speed and displacement, which describes the motion state of the main vehicle at a certain moment.

[0059] The motion set defines five uniformly accelerated motions: rapid acceleration, rapid deceleration, constant speed, gradual acceleration, and gradual deceleration, which are the possible actions of the main vehicle.

[0060] The reward function takes into account multiple factors, such as the minimum distance to dynamic obstacles in the current state, the minimum distance to static obstacles, and the vehicle speed in the current state, to evaluate the quality of the action sequence. The higher the reward value, the better the action sequence.

[0061] For the Monte Carlo tree search process, a branch reduction operation is introduced when expanding the search tree, generating only feasible child nodes. Feasible child nodes must satisfy the following constraints: lateral acceleration does not exceed a threshold (curvature is a function of displacement), speed does not exceed the speed limit (speed limit is a function of displacement), there is no collision with obstacles, and the distance to the vehicle in front in the final state meets the safe distance requirement. The generated motion sequences are randomly simulated to predict the motion state of the main vehicle under different actions. For example, when the displacement coordinate exceeds 30m and enters a curve (curvature of 1 / 50m), the process of decelerating to 12m / s upon entering the curve is simulated to avoid exceeding the lateral acceleration limit.

[0062] Obstacle trajectories are introduced in the following three ways: For non-interactive trajectories, the predicted obstacle trajectories are used, and the space occupied by the obstacles is reflected in the ST diagram as a static constraint.

[0063] Based on the interactivity of the car-following model, after each action of the master vehicle, the future state of the environment vehicle is not obtained through static trajectory prediction, but is calculated and updated in real time according to the current behavior of the master vehicle. The MCTS extended node not only contains the state of the master vehicle, but also the updated state of the environment vehicle, to reflect interactivity. Specifically, the master vehicle first performs an action (such as acceleration or deceleration), and the environment vehicle, based on the master vehicle's behavior, uses the car-following algorithm to calculate its future speed and position, thereby generating a new state node. For example... Figure 3 The interactive diagram illustrates that when the primary vehicle A chooses to insert its action, the environment vehicle B maintains the predicted straight-line trajectory but decelerates based on a following model (which can use the Intelligent Driver Model), generating a new state node. The primary vehicle's state is calculated from its behavior and current state using kinematic formulas; the environment vehicle's state is calculated by referencing the primary vehicle's behavior, its old and new states, and the dynamic and static traffic environment. This approach is the recommended method.

[0064] The Stackelberg Driver Model (SLM) employs a multi-vehicle cooperative approach, where the primary vehicle acts as the leader and environmental vehicles as followers. Each follower selects the optimal response strategy based on the leader's actions. The environmental vehicles' responses are not predefined but dynamically calculated to achieve speed planning for multi-vehicle cooperation. The model output is derived from the vehicle trajectories, and the corresponding longitudinal action (acceleration) sequence is derived from the model's output speed curve, generating an initial branch before the MCTS search begins. If the action sequence from the model corresponds to a high reward, MCTS tends to converge quickly and produce results similar to the model's output. If the action sequence from the model corresponds to a low reward, MCTS tends to explore new directions and make decisions different from the model to improve search efficiency and accuracy.

[0065] For velocity planning based on optimization algorithms, after obtaining the coarse velocity planning result from MCTS, optimization algorithms such as QP / IPOPT are used to solve for a smooth velocity curve within the space found by MCTS in the ST diagram. Specifically: For the cost function, factors such as deviation from the target position and cumulative acceleration are considered to ensure the smoothness and accuracy of the speed curve, so that the vehicle can reach the target position smoothly and reduce passenger discomfort.

[0066] The constraints include initial motion state, final motion state, displacement constraints, and kinematic constraints. Among them, the displacement constraints are derived from the output of MCTS combined with the lower and upper limits found in the ST plot, to ensure that the optimized velocity curve is within the feasible range.

[0067] In the embodiments of this specification, in the process of determining the specified scene in which the current vehicle is located, traffic sign information at a preset distance in front of the current vehicle can be obtained first; then, the specified scene can be determined based on the driving information, the motion information and the traffic sign information.

[0068] It should be noted that traffic sign information within a preset distance in front of the vehicle can be obtained through onboard forward vision sensors (such as cameras) or vehicle-to-everything (V2X) communication modules. This information includes, but is not limited to, traffic light status, road sign content, and road marking types. Based on the current vehicle driving information obtained in S101, the traffic participant movement information obtained in S102, and the aforementioned traffic sign information, a scene classification model is used for multi-source information fusion analysis. By comprehensively judging the vehicle's position, the behavior of surrounding traffic participants, and traffic rule requirements, the specified scene type (such as traffic light intersection, pedestrian crossing, speed limit section, etc.) of the current vehicle is accurately determined.

[0069] It should be noted that this application obtains traffic sign information within a preset distance ahead of the vehicle and integrates it with the vehicle's driving status and the movement information of surrounding participants to accurately identify the specific scenario (such as intersections, speed-limited zones, pedestrian crossings, etc.) in which the vehicle is currently located. This mechanism allows the system to move beyond relying on single geometric or kinematic information and instead incorporate high-level semantic rules, thereby transforming abstract traffic regulations and road constraints into calculable and operable scenario types. Based on this specified scenario, the system can accurately invoke the corresponding displacement-time graph constraint range, ensuring that subsequent action search and speed optimization strictly adhere to traffic rules and safety criteria. This effectively avoids decision-making biases caused by scenario misjudgment or rule omissions, ultimately enhancing the compliance, environmental adaptability, and overall system reliability of the speed planning results.

[0070] Optionally, in the process of inputting the constraint range into a pre-built action search model to obtain the optimal action sequence of the current vehicle, within a pre-set decision time range, the longitudinal motion behavior of the current vehicle is discretized into multiple acceleration actions; the multiple acceleration actions are input into the action search model constructed by the Monte Carlo tree search algorithm, and search optimization is performed within the constraint range to obtain the optimal acceleration action sequence that satisfies all constraint conditions.

[0071] It should be noted that, within a pre-defined decision-making timeframe, the current longitudinal motion of the vehicle is discretized into a finite set of acceleration actions. These actions cover basic driving operations such as rapid acceleration, rapid deceleration, constant speed, gradual acceleration, and gradual deceleration, forming a set of discrete actions available for search. The discretized acceleration actions are then input into an action search model constructed using a Monte Carlo tree search algorithm. This model performs search optimization within the displacement-time graph constraints defined in S105. Through iterative execution of selection, expansion, simulation, and backtracking processes, the performance of different action sequences is evaluated under constraints (such as speed limits, collision avoidance, and safe following distance), ultimately outputting an acceleration action sequence that satisfies all constraints and has the optimal overall evaluation.

[0072] It should be noted that this application significantly improves the decision-making quality and efficiency of velocity planning in complex dynamic environments by discretizing longitudinal motion behavior into multiple acceleration actions and using the Monte Carlo Tree Search (MCTS) algorithm to search within a predefined constraint range. This method transforms the continuous decision space into a finite, manageable set of actions through discretization, providing a structured search foundation for MCTS and effectively overcoming the high computational complexity and susceptibility to local optima inherent in traditional optimization algorithms in high-dimensional continuous spaces. Simultaneously, by directly embedding the constraint range extracted from the displacement-time graph into the MCTS search process, it ensures that the algorithm explores only within the feasible solution space that meets safety and rule requirements, greatly reducing invalid search paths and enhancing the targeting and safety of the decision. Ultimately, MCTS can output a globally superior acceleration action sequence within a preset decision time, thereby providing a high-quality initial solution for subsequent optimization algorithms and comprehensively improving the reliability, real-time performance, and adaptability to dynamic interactive scenarios of the velocity planning results.

[0073] Optionally, when constructing the action search model, the state set, action set, and reward function of the action search model are determined; wherein, the state set includes the current speed and displacement of the vehicle; the action set includes rapid acceleration, rapid deceleration, constant speed, gradual acceleration, and gradual deceleration; the reward function is constructed based on the minimum distance to the traffic participant and the first speed information, and is used to evaluate the quality of the action sequence, with a higher reward value indicating a better action sequence.

[0074] It should be noted that when constructing the action search model, the core components of the model are first defined. The model's state set is determined, consisting of the current vehicle's speed and displacement; the model's action set is determined, including predefined longitudinal control actions such as rapid acceleration, rapid deceleration, constant speed, gradual acceleration, and gradual deceleration; simultaneously, a reward function is constructed, which uses the minimum time distance between the current vehicle and traffic participants, as well as the current vehicle's initial speed information, as core input parameters. This function is used to quantitatively evaluate the comprehensive performance of different action sequences during the search process; a higher reward value indicates a higher priority for the corresponding action sequence.

[0075] It should be noted that this application significantly improves the scientific rigor, safety, and efficiency of the speed planning decision-making process by constructing an action search model with clearly defined states, a reasonable action space, and a targeted reward function. This model abstracts the vehicle state into a set containing speed and displacement, comprehensively representing the vehicle's real-time motion characteristics. The action set covers typical longitudinal driving behaviors such as rapid acceleration, rapid deceleration, constant speed, gradual acceleration, and gradual deceleration, ensuring the coverage of the decision-making process while reducing search complexity through discretization. Crucially, the reward function is constructed based on the minimum distance to traffic participants and the vehicle's speed, effectively transforming the requirements for maintaining safe distances and smooth driving into quantifiable optimization objectives. This guides the Monte Carlo tree search algorithm to quickly select acceleration action sequences that ensure safe distances while also considering driving efficiency and comfort within constraints. This mechanism not only avoids the blindness of unconstrained random searches but also strengthens the targeted response to key risk factors in dynamic interaction scenarios, ultimately significantly enhancing the overall quality, reliability, and environmental adaptability of the speed planning results.

[0076] Optionally, when inputting the constraint range into a pre-built action search model to obtain the optimal action sequence of the current vehicle, a branch reduction operation is performed when the Monte Carlo tree search algorithm expands the search tree to generate candidate action sequences that meet the constraints. The constraints include: the first speed does not exceed a preset speed limit, there is no collision with the traffic participants, and the distance to the forward traffic participants meets the safe distance. The candidate action sequences are randomly simulated to predict the motion state of the current vehicle under different candidate action sequences, and the optimal action sequence of the current vehicle is determined based on the motion state.

[0077] It should be noted that during the Monte Carlo tree search algorithm's search tree expansion, a branch reduction operation is performed simultaneously. This operation, based on the constraints determined in S105, removes branches that violate the following constraints in real time: the current vehicle's initial speed exceeds a preset speed limit, there is a risk of collision with the predicted trajectory of traffic participants, or the distance to the forward traffic participant is below a safe distance threshold. This operation generates a set of candidate action sequences that fully satisfy the constraints. The candidate action sequences generated by the branch reduction operation are then subjected to random simulation. By predicting the future motion state (including changes in position and speed) of the current vehicle when executing different candidate action sequences, and comprehensively evaluating the interaction between these states and the predicted trajectory of traffic participants, the optimal action sequence is finally determined as the output.

[0078] It should be noted that this application significantly improves the efficiency, safety, and decision reliability of action sequence search by deeply integrating constraints into the expansion and simulation stages of the Monte Carlo Tree Search algorithm. This method actively performs a branch reduction operation when expanding the search tree, strictly filtering action branches based on preset key constraints such as speed limits, collision avoidance, and safe distances. This fundamentally eliminates a large number of invalid actions that violate traffic rules or pose safety hazards, greatly compressing the search space and avoiding the waste of computational resources and decreased timeliness caused by blindly exploring all possible actions in traditional MCTS. Subsequently, by randomly simulating the candidate action sequences generated after branch reduction, the future motion state of the vehicle under different action sequences is accurately predicted. This allows for the evaluation of sequence quality within the feasible solution space based on actual motion effects rather than simply heuristic rules. This ensures that the final optimal action sequence not only theoretically satisfies all hard constraints but also exhibits good dynamic adaptability and practical feasibility in the simulation environment, thereby comprehensively enhancing the real-time performance, safety, and overall decision quality of the speed planning results.

[0079] Optionally, the random simulation of the candidate action sequence includes at least one of the following modes: using a non-interactive trajectory mode, where the traffic participant's predicted trajectory is used as a static constraint in the displacement-time graph; or, using an interactive mode based on a vehicle-following model, where the traffic participant's state is updated in real-time at each extended node of the Monte Carlo tree search based on the current vehicle's action; wherein the traffic participant calculates its future motion state based on the vehicle-following model and the current vehicle's behavior; or, using a multi-vehicle cooperative mode, where the current vehicle is the dominant player and the traffic participant is the follower, dynamically calculating the traffic participant's optimal response strategy based on the current vehicle's action.

[0080] It should be noted that when performing random simulations, at least one of the following modes can be selected for execution depending on the scenario requirements: In the non-interactive trajectory mode, the trajectory information of traffic participants predicted in S103 is used as a static, time-invariant constraint boundary, directly mapped and reflected in the displacement-time graph generated in S104. The simulation process is evaluated based on this static constraint environment.

[0081] The interactive mode based on the vehicle-following model can be adopted. In each expansion node of the Monte Carlo tree search algorithm, based on the assumed actions of the current vehicle, the predefined vehicle-following model (such as the Intelligent Driver Model, IDM) is used to calculate in real time the possible state changes of traffic participants, thereby dynamically updating their future motion state for simulation.

[0082] In a multi-vehicle cooperative mode, the current vehicle is designated as the dominant vehicle, while other traffic participants are designated as followers. During the simulation, based on the assumed actions of the current vehicle, the optimal response strategies of the traffic participants are dynamically calculated using game theory or a cooperative decision-making model, and their states are updated accordingly for joint simulation.

[0083] It should be noted that this application significantly improves the decision-making intelligence, safety, and adaptability of autonomous driving speed planning in complex interactive scenarios by introducing a multi-mode stochastic simulation mechanism. This method provides three simulation modes: non-interactive trajectory, interactive mode based on a following model, and multi-vehicle cooperative mode. These modes can be flexibly selected or combined according to actual scenario requirements, thus systematically solving the problems of single environmental modeling and lagging interactive feedback in traditional methods. The non-interactive mode treats traffic participants as static constraints, providing a basic safety boundary guarantee for decision-making. The interactive mode updates the status of surrounding vehicles in real time through a following model, enabling the simulation process to dynamically reflect the impact of the vehicle's behavior on other vehicles, enhancing the accuracy of trajectory prediction and the coupling of decision-making. The multi-vehicle cooperative mode further treats the vehicle as the dominant player, simulating the optimal response strategy of traffic participants, achieving efficient modeling of cooperative driving scenarios. This multi-level, configurable simulation framework allows the Monte Carlo tree search algorithm to evaluate the long-term effects of action sequences in a more realistic and dynamic environment, thereby generating optimal action sequences that meet immediate safety constraints while possessing good interactivity and foresight, comprehensively improving the reliability, robustness, and scenario generalization ability of the speed planning system.

[0084] Optionally, when inputting the optimal action sequence into a pre-set optimization algorithm to plan the speed curve of the current vehicle, the optimization algorithm can be used to solve for a smooth speed curve based on the optimal action sequence and the constraint range in the displacement-time graph; wherein, the cost function of the optimization algorithm considers the deviation from the path information and the cumulative value of acceleration; the constraints of the optimization algorithm include initial motion state, final motion state, displacement constraints and kinematic constraints, and the displacement constraints are derived from the upper and lower limits of the displacement corresponding to the optimal action sequence.

[0085] It should be noted that the optimal action sequence obtained in S106 and the constraint range in the displacement-time graph generated in S104 are used as inputs and fed into a pre-defined optimization algorithm. This optimization algorithm is used to solve for a smooth velocity curve. The cost function of the algorithm mainly considers two factors: the deviation between the actual trajectory of the current vehicle and the path information obtained in S101, and the cumulative value of acceleration over the entire planning period (used to measure comfort). The constraints of the algorithm include: using the current real-time motion state of the vehicle as the initial state of optimization, using the expected target state as the final state, strictly adhering to the upper and lower limits of displacement corresponding to the optimal action sequence (as displacement constraints), and satisfying the inherent kinematic constraints of the vehicle (such as the physical limits of acceleration and jerk).

[0086] It should be noted that this application combines the optimal action sequence output by Monte Carlo tree search with the optimization algorithm, and introduces a cost function with clear engineering significance and multi-dimensional constraints, achieving a high-quality transformation from discrete action sequences to continuous and smooth velocity curves, significantly improving the comfort, accuracy, and feasibility of velocity planning. This method uses the upper and lower limits of displacement provided by the optimal action sequence as core displacement constraints, ensuring that the optimization process always remains within a safe and feasible solution space. Simultaneously, the cost function of the optimization algorithm considers both the tracking deviation on the predetermined path and the cumulative acceleration value, effectively balancing the two key performance indicators of driving accuracy and ride comfort. Furthermore, the strict adherence to the initial and final motion states and vehicle kinematic constraints ensures that the final generated velocity curve not only fully inherits the safety and rationality of the front-end decision-making module, but also achieves a deep integration of trajectory smoothness, power adaptability, and control precision in the continuous domain, thereby comprehensively enhancing the overall reliability, ride quality, and adaptability to complex dynamic environments when the autonomous driving system executes velocity planning.

[0087] It should be noted that the hysteresis mechanism in the longitudinal state machine / decision-making process can have its entry and exit conditions adjusted according to the actual scenario. For example, in traffic light traffic decisions, the conditions for entering the stop-and-wait state can be stricter or more lenient, as long as frequent decision fluctuations are avoided. In the coarse speed planning based on MCTS, in addition to the five uniform acceleration movements (rapid acceleration, rapid deceleration, constant speed, gradual acceleration, and gradual deceleration), other types of acceleration movements can be added, such as more refined acceleration gradations, as long as they meet the requirements of discretized movements and cover the main driving behaviors. In the speed planning based on optimization algorithms, in addition to QP / IPOPT, other types of optimization algorithms can be used, such as the MPC algorithm, as long as a smooth speed curve can be solved within the space defined by MCTS. In the obstacle trajectory introduction method, the car-following algorithm used based on the interactivity of the car-following model can be replaced with other types of car-following models, such as car-following models with different parameter settings or car-following algorithms based on other principles, as long as the function of updating the environmental vehicle state in real time according to the behavior of the main vehicle can be achieved.

[0088] It should be noted that a four-step algorithm architecture—"relevant traffic participant search → longitudinal state machine / decision → coarse speed planning based on MCTS → speed planning based on optimization algorithm"—is adopted, realizing a complete process from environmental perception to speed curve generation. A rule-based hysteresis mechanism is introduced in the longitudinal state machine / decision to avoid frequent inter-frame decision fluctuations; and the scene decision results are transformed into various constraint forms in the ST graph, effectively limiting the search range of MCTS. In the coarse speed planning based on MCTS, a reasonable Markov decision process (state set, action set, reward function) is defined, and a branch reduction operation is introduced during the expansion process to generate only feasible child nodes; simultaneously, obstacle trajectories are introduced in multiple ways (non-interactive, interactive based on a car-following model, multi-vehicle collaboration), considering the interactivity of traffic participants; in addition, the self-vehicle trajectory output by the model is introduced as the initial branch, improving the search efficiency of MCTS. The speed planning based on the optimization algorithm is based on the coarse results of MCTS, solving for a smooth speed curve within a specific space, balancing the optimality and smoothness of speed planning.

[0089] Figure 4 This is a schematic diagram of the structure of a speed planning device for autonomous driving of a vehicle, including: an information acquisition unit 401, an information determination unit 402, a prediction unit 403, a generation unit 404, a range determination unit 405, a sequence determination unit 406, and a speed planning unit 407.

[0090] The information acquisition unit 401 acquires the driving information of the current vehicle during the current autonomous driving process. The driving information includes path information, first location information and first speed information. Information determination unit 402 determines the motion information of traffic participants based on the driving information, the motion information including second position information, second speed information and motion direction information; The prediction unit 403 predicts the trajectory information of the traffic participant based on the motion information; The generation unit 404 generates a displacement-time map of the current vehicle based on the trajectory information. The displacement-time map represents the displacement range of the traffic participant relative to the current vehicle at different times. The range determination unit 405 determines the specified scenario in which the current vehicle is located and determines the constraint range in the displacement-time graph corresponding to the specified scenario; The sequence determination unit 406 inputs the constraint range into a pre-built action search model to obtain the optimal action sequence of the current vehicle; The speed planning unit 407 inputs the optimal action sequence into a pre-set optimization algorithm to plan the speed curve of the current vehicle.

[0091] Figure 5 A schematic diagram of the structure of a speed planning device for autonomous driving of a vehicle, including: At least one processor and bus; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: During the current autonomous driving process of the vehicle, the driving information of the current vehicle is acquired, including path information, first location information and first speed information; Based on the driving information, the motion information of the traffic participants is determined, including second position information, second speed information, and motion direction information; Based on the motion information, predict the trajectory information of the traffic participants; Based on the trajectory information, a displacement-time map of the current vehicle is generated, which represents the displacement range of the traffic participant relative to the current vehicle at different times. Determine the specified scenario in which the current vehicle is located, and determine the constraint range in the displacement-time graph corresponding to the specified scenario; The constraint range is input into a pre-built action search model to obtain the optimal action sequence for the current vehicle; The optimal action sequence is input into a pre-set optimization algorithm to plan the speed curve of the current vehicle.

[0092] Corresponding to the above scheme, Figure 6 The diagram shows the structure of a vehicle, which includes a memory 601 and a processor 602. The memory 601 stores executable program code 6011, and the processor 602 is used to call and execute the executable program code 6011 to perform a mode switching method for a hybrid vehicle.

[0093] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0094] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0095] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0096] In the embodiments provided in this application, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0097] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0098] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The aforementioned units can be implemented in hardware or software.

[0099] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0100] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A speed planning method for autonomous driving of vehicles, characterized in that, The method includes: During the current autonomous driving process of the vehicle, the driving information of the current vehicle is acquired, including path information, first location information and first speed information; Based on the driving information, the motion information of the traffic participants is determined, including second position information, second speed information, and motion direction information; Based on the motion information, predict the trajectory information of the traffic participants; Based on the trajectory information, a displacement-time map of the current vehicle is generated, which represents the displacement range of the traffic participant relative to the current vehicle at different times. Determine the specified scenario in which the current vehicle is located, and determine the constraint range in the displacement-time graph corresponding to the specified scenario; The constraint range is input into a pre-built action search model to obtain the optimal action sequence for the current vehicle; The optimal action sequence is input into a pre-set optimization algorithm to plan the speed curve of the current vehicle.

2. The method according to claim 1, characterized in that, Determining the specified scenario in which the current vehicle is located includes: Obtain traffic sign information at a preset distance ahead of the current vehicle; The designated scenario is determined based on the driving information, the motion information, and the traffic sign information.

3. The method according to claim 1, characterized in that, The step of inputting the constraint range into a pre-built action search model to obtain the optimal action sequence for the current vehicle includes: Within a pre-defined decision time range, the longitudinal motion behavior of the current vehicle is discretized into multiple acceleration actions; Multiple acceleration actions are input into an action search model constructed by a Monte Carlo tree search algorithm, and search optimization is performed within the constraints to obtain the optimal acceleration action sequence that satisfies all constraints.

4. The method according to claim 3, characterized in that, The construction of the action search model includes: Determine the state set, action set, and reward function of the action search model; wherein, The state set includes the current speed and displacement of the vehicle; The set of actions includes rapid acceleration, rapid deceleration, constant speed, gradual acceleration, and gradual deceleration; The reward function is constructed based on the minimum distance to the traffic participant and the first speed information, and is used to evaluate the quality of the action sequence. The higher the reward value, the better the corresponding action sequence.

5. The method according to claim 3, characterized in that, The step of inputting the constraint range into a pre-built action search model to obtain the optimal action sequence for the current vehicle includes: When the Monte Carlo tree search algorithm expands the search tree, a branch reduction operation is performed to generate a sequence of candidate actions that meet the constraints. The constraints include: the first speed does not exceed a preset speed limit, there is no collision with the traffic participant, and the distance to the forward traffic participant meets the safe distance. The candidate action sequences are randomly simulated to predict the motion state of the current vehicle under different candidate action sequences, and the optimal action sequence of the current vehicle is determined based on the motion state.

6. The method according to claim 5, characterized in that, The random simulation of the candidate action sequence includes at least one of the following modes: A non-interactive trajectory mode is adopted, in which the trajectories of the traffic participants based on the prediction are reflected as static constraints in the displacement-time graph; Alternatively, an interactive mode based on a vehicle-following model can be adopted, in which the state of the traffic participant is updated in real time according to the current vehicle's action at each extended node of the Monte Carlo tree search; wherein the traffic participant calculates its future motion state based on the behavior of the current vehicle using the vehicle-following model. Alternatively, a multi-vehicle cooperative mode can be adopted, in which the current vehicle plays the leading role and the traffic participants play the following role, and the optimal response strategy of the traffic participants is dynamically calculated based on the actions of the current vehicle.

7. The method according to claim 1, characterized in that, The step of inputting the optimal action sequence into a pre-set optimization algorithm to plan the speed curve of the current vehicle includes: Based on the optimal action sequence and the constraint range in the displacement-time graph, the optimization algorithm is used to solve for the smooth velocity curve. The cost function of the optimization algorithm takes into account the deviation from the path information and the cumulative value of acceleration. The constraints of the optimization algorithm include initial motion state, final motion state, displacement constraints, and kinematic constraints. The displacement constraints are derived from the upper and lower limits of the displacement corresponding to the optimal action sequence.

8. A speed planning device for autonomous driving of vehicles, characterized in that, include: The information acquisition unit acquires the driving information of the current vehicle during the current autonomous driving process. The driving information includes path information, first location information, and first speed information. The information determination unit determines the motion information of the traffic participant based on the driving information, the motion information including second position information, second speed information and motion direction information; The prediction unit predicts the trajectory information of the traffic participant based on the motion information; The generation unit generates a displacement-time map of the current vehicle based on the trajectory information. The displacement-time map represents the displacement range of the traffic participant relative to the current vehicle at different times. The range determination unit determines the specified scenario in which the current vehicle is located and determines the constraint range in the displacement-time graph corresponding to the specified scenario; The sequence determination unit inputs the constraint range into a pre-built action search model to obtain the optimal action sequence of the current vehicle; The speed planning unit inputs the optimal action sequence into a pre-set optimization algorithm to plan the speed curve of the current vehicle.

9. A speed planning device for autonomous driving of vehicles, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: During the current autonomous driving process of the vehicle, the driving information of the current vehicle is acquired, including path information, first location information and first speed information; Based on the driving information, the motion information of the traffic participants is determined, including second position information, second speed information, and motion direction information; Based on the motion information, predict the trajectory information of the traffic participants; Based on the trajectory information, a displacement-time map of the current vehicle is generated, which represents the displacement range of the traffic participant relative to the current vehicle at different times. Determine the specified scenario in which the current vehicle is located, and determine the constraint range in the displacement-time graph corresponding to the specified scenario; The constraint range is input into a pre-built action search model to obtain the optimal action sequence for the current vehicle; The optimal action sequence is input into a pre-set optimization algorithm to plan the speed curve of the current vehicle.

10. A vehicle, characterized in that, Including the speed planning method for autonomous driving of vehicles as described in any one of claims 1 to 7.

Images