Heavy truck decision-making method, system and equipment based on Frenet grid and layered dynamic game, and medium

Abstract

The embodiment of the invention provides a heavy truck decision-making method, system and equipment based on a Frenet grid and a hierarchical dynamic game and a medium, and belongs to the field of automatic driving. The method comprises the following steps: dispersing a road space around a vehicle into Frenet grids; quantifying lane traffic risks, inter-vehicle interaction risks and traffic rule violation risks on the basis of characteristics of codes in the Frenet grids, and generating a dynamic risk thermodynamic diagram through space-time convolution; and constructing a non-cooperative game model containing the vehicle and other traffic participants, taking the dynamic risk thermodynamic diagram as a constraint condition of a strategy space, introducing a traffic rule penalty term into a game revenue function, and solving Nash equilibrium to obtain an optimal driving strategy of the vehicle. In complex dynamic traffic, safety, compliance and traffic efficiency are considered at the same time, and an optimal driving strategy is generated.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and more specifically to a heavy-duty truck decision-making method, system, device, and medium based on Frenet grid and hierarchical dynamic game. Background Technology

[0002] Currently, Level 2 and above autonomous driving systems face three major technical bottlenecks in complex dynamic scenarios (such as urban roads, highways, and intersections): First, insufficient interaction modeling: traditional methods struggle to accurately represent the dynamic interactions between vehicles and between vehicles and pedestrians / the environment, leading to insufficient decision robustness and a tendency for misjudgments or conflicts in dense traffic or unexpected situations. Second, poor policy generalization: rule-based or fixed-policy decision-making systems struggle to adapt to diverse traffic scenarios (such as urban roads, highways, and intersections), resulting in inefficient or even ineffective lane-changing and following decisions. Third, poor rule compatibility: existing game theory methods often result in decisions that violate traffic rules due to conflicts between individual optimality and collective interests, or rigid rule embedding methods, such as mandatory adherence to external constraints, affecting system safety and regulatory compliance.

[0003] Existing intelligent driving decision-making methods, such as rule-based decision-making and traditional reinforcement learning, suffer from the following problems in complex and dynamic traffic scenarios: Rule-based decision-making methods generate driving strategies through pre-defined if-then rules (such as following distance thresholds and lane-changing conditions). Typical applications include finite state machines (FSMs) and decision trees. However, they have several drawbacks: First, limited scenario coverage: rules require manual design, making it difficult to cover all long-tail scenarios (such as vehicles cutting in line or changing lanes in construction zones), resulting in poor generalization ability. Second, lack of interaction modeling: only considering the vehicle's own state and ignoring the intentions of other traffic participants leads to strategy failure in game-theoretic scenarios (such as the inability to negotiate when merging into congested traffic). Third, rule conflicts: multiple rules running in parallel may cause contradictions (such as the conflict between "maintaining a safe following distance" and "avoiding crossing solid lines"), requiring complex priority arbitration and incurring significant computational burden.

[0004] Traditional reinforcement learning methods drive agents to learn strategies through reward functions, such as DQN and PPO algorithms. However, they have several drawbacks: first, low training efficiency: requiring massive amounts of interactive data, and the sparse reward problem leads to convergence difficulties (e.g., successful lane changes only take effect at the final stage); second, poor policy interpretability: black-box models make it difficult to verify safety, and "shortcut strategies" may be learned (e.g., crossing solid lines to overtake and shorten the path); and third, weak rule compatibility: reward functions require manually designed rules (e.g., penalties for running red lights), but weight adjustment is difficult, leading to rule avoidance (e.g., minor line crossings go unpunished but occur frequently). Summary of the Invention

[0005] The purpose of this invention is to provide a heavy-duty truck decision-making method, system, device, and medium based on Frenet grid and hierarchical dynamic game. By constructing a Frenet grid, a standardized environmental representation is provided for decision-making. Dynamic risk heatmaps are used to achieve refined real-time assessment of scenario risks. On this basis, a hierarchical game model integrating rule constraints is constructed, which can generate the optimal driving strategy in complex dynamic traffic while taking into account safety, compliance, and traffic efficiency.

[0006] To achieve the above objectives, embodiments of the present invention provide a heavy-duty truck decision-making method based on Frenet grids and hierarchical dynamic game theory, comprising: The road space surrounding the vehicle is discretized into a Frenet grid, wherein the Frenet grid comprises multiple cells, and each cell is encoded with multidimensional features; Based on the features encoded in the Frenet mesh, lane traffic risk, vehicle-to-vehicle interaction risk, and traffic rule violation risk are quantified, and a dynamic risk heatmap is generated through spatiotemporal convolution. The dynamic risk heatmap is used to characterize the risk level at different spatiotemporal locations. A non-cooperative game model involving the vehicle and other traffic participants is constructed. The dynamic risk heatmap is used as a constraint on the strategy space, and traffic rule penalty terms are introduced into the game payoff function. The optimal driving strategy of the vehicle is obtained by solving the Nash equilibrium.

[0007] Optionally, the road space surrounding the vehicle can be discretized into a Frenet mesh, including: Centered on the vehicle, the longitudinal and transverse road space is discretized into a grid matrix, where the longitudinal direction represents the road direction and the transverse direction represents the lane direction; The information of traffic participants and obstacles is projected onto the Frenet grid matrix to obtain the kinematic characteristics of traffic participants and obstacles under the Frenet grid matrix, and the grid occupancy rate of traffic participants and obstacles is calculated. Based on the grid occupancy rates of the traffic participants and obstacles, a multidimensional feature vector is filled into each cell of the Frenet grid matrix. The multidimensional feature vector includes at least: object type, boundary, velocity vector, acceleration vector, occupancy rate, and game marker.

[0008] Optionally, a dynamic risk heatmap can be generated through spatiotemporal convolution, including: Calculate the lane flow index and lane speed standard deviation to assess lane-level traffic risk; Based on the gap and speed difference between vehicles in front and behind, lane-changing feasibility indicators are calculated to assess the risk of interaction between vehicles. Based on a set of traffic rules, potential rule violations are assessed to determine the risk of traffic rule violations. Design a separable spatiotemporal convolution kernel, wherein the spatial convolution kernel is a Gaussian function whose standard deviation is related to the vehicle speed, and the temporal convolution kernel is a multi-frame sliding window weight vector; The lane-level traffic risks, vehicle-to-vehicle interaction risks, and traffic rule violation risks are weighted and fused together, and then convolutional operations are performed using the spatiotemporal convolution kernel to output the final dynamic risk heatmap.

[0009] Optionally, the lane flow index can be calculated using the following formula: ; In the formula, It is a lane in the grid The number of vehicles inside, It is the first The speed of the car It is the vehicle type coefficient. It is a lane width, It is the free flow velocity. This is an adjustable coefficient. The slope angle is denoted by .

[0010] Optionally, the lane speed standard deviation can be calculated using the following formula: ; In the formula, It is a lane in the grid The number of vehicles inside, It is the first The speed of the car It is a lane The average speed.

[0011] Optionally, a non-cooperative game model is constructed, including the vehicle itself and other traffic participants. The dynamic risk heatmap is used as a constraint on the strategy space, and traffic rule penalties are introduced into the game payoff function. The optimal driving strategy for the vehicle is obtained by solving for the Nash equilibrium, including: Define a set of game participants and define a strategy space for each game participant. The strategy space consists of a combination of lateral and vertical strategies. Lateral strategies include lane keeping, left lane change, and right lane change. Vertical strategies include deceleration, following, and acceleration. Construct a revenue function, wherein the revenue function includes the vehicle's own revenue and the revenue from interactions with other participants; By introducing a rule penalty term into the profit function, a modified profit function is obtained. Based on the dynamic risk heatmap, areas with risk values ​​exceeding the high-risk threshold are marked as infeasible strategy spaces, so as to perform constraint pruning on the strategy space of each participant. An iterative strategy is adopted, combining hot-start initialization and constraint pruning, to solve the Nash equilibrium solution of the non-cooperative game model, which is then used as the optimal driving strategy for this vehicle.

[0012] Optionally, the constrained pruning includes: Traverse the participants' strategy space, eliminate strategies marked as violating traffic rules by the indicator function in the payoff function; and predict the trajectory of the remaining strategies. If the predicted trajectory crosses the infeasible strategy space, then eliminate the strategy. If a participant's available strategy space is empty after pruning, the high-risk threshold is adjusted and the selection is re-screened to ensure that at least one feasible strategy is retained.

[0013] Secondly, the present invention also provides a heavy-duty truck decision-making system based on Frenet grid and hierarchical dynamic game, comprising: The Frenet mesh building module is used to discretize the road space around the vehicle into a Frenet mesh, wherein the Frenet mesh includes multiple cells and encodes multidimensional features for each cell; The risk quantification module is used to quantify lane traffic risk, vehicle-to-vehicle interaction risk, and traffic rule violation risk based on the features encoded in the Frenet grid, and to generate a dynamic risk heat map through spatiotemporal convolution, wherein the dynamic risk heat map is used to characterize the risk level at different spatiotemporal locations. The game optimization module is used to construct a non-cooperative game model that includes the vehicle itself and other traffic participants. The dynamic risk heat map is used as a constraint on the strategy space, and traffic rule penalty terms are introduced into the game payoff function. The optimal driving strategy of the vehicle is obtained by solving the Nash equilibrium.

[0014] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the heavy truck decision-making method based on Frenet grid and hierarchical dynamic game described above.

[0015] Fourthly, the present invention also provides a storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of the heavy truck decision-making method based on Frenet grid and hierarchical dynamic game described above.

[0016] The above technical solution discretizes the continuous and complex traffic environment into a structured, multi-feature encoded grid representation by constructing a Frenet grid, providing a high-precision and standardized data foundation for subsequent processing and improving environmental modeling capabilities and the input quality of the decision-making system. Utilizing adaptive spatiotemporal convolution kernels, multi-dimensional risk sources such as lane flow, vehicle interaction, and traffic rules are integrated to generate a dynamic risk heatmap, enabling refined and real-time quantitative assessment of scenario risks. Introducing the risk heatmap as a hard constraint into a hierarchical non-cooperative game framework, combined with rule-based penalties, allows the Nash equilibrium solution to simultaneously optimize driving efficiency, avoid high-risk areas, and naturally comply with traffic rules. This generates intelligent driving decisions that combine safety, compliance, efficiency, and strong interactive adaptability in complex dynamic scenarios.

[0017] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a heavy truck decision-making method based on Frenet grid and hierarchical dynamic game provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a heavy truck decision-making system based on Frenet grid and hierarchical dynamic game provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] Various embodiments of this disclosure will be described more fully in the following detailed description. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0020] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions or operations and do not limit the addition of one or more functions or operations. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a specific feature, number, step, operation, or combination of the foregoing and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, or combinations of the foregoing, or the possibility of adding one or more features, numbers, steps, operations, or combinations of the foregoing.

[0021] In various embodiments of this disclosure, the expression "or" or "at least one of A and / or B" includes any combination or all combinations of the words listed simultaneously. For example, the expression "A or B" or "at least one of A and / or B" may include A, may include B, or may include both A and B.

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] See Figure 1 The diagram shows a flowchart of a heavy-duty truck decision-making method based on Frenet grid and hierarchical dynamic game in a specific embodiment, including the following execution steps: Step 100: Discretize the road space around the vehicle into a Frenet mesh.

[0024] The Frenet grid comprises multiple cells, and each cell is encoded with multidimensional features.

[0025] Specifically, when executing step 100, the following steps can be performed: S1000: Centered on the vehicle, the longitudinal and lateral road space is discretized into a grid matrix, where the longitudinal direction represents the direction along the road and the lateral direction represents the direction of the lane.

[0026] S1001: Project traffic participants and obstacle information onto the Frenet grid matrix, obtain the kinematic characteristics of traffic participants and obstacles under the Frenet grid matrix, and calculate the grid occupancy rate of traffic participants and obstacles.

[0027] Multidimensional feature encoding of networks that integrate kinematic and geometric features compresses continuous, heterogeneous physical information into a standardized, discrete grid feature vector, enabling subsequent risk assessment and game theory to be performed based on efficient multidimensional array operations.

[0028] S1002: Based on the grid occupancy rate of the traffic participants and obstacles, fill each cell in the Frenet grid matrix with a multidimensional feature vector, wherein the multidimensional feature vector includes at least: object type, boundary, velocity vector, acceleration vector, occupancy rate, and game marker.

[0029] In one specific implementation, the Frenet grid can be constructed through the following steps: Step 1: Discretize the area around the vehicle into a 5×n grid matrix. The longitudinal area range (unit: meters) and the number of longitudinal grids n can be adjusted as needed. The following description assumes n=5. The lane where the vehicle is located is the horizontal coordinate 3 of the grid matrix. The horizontal coordinates of the grid matrix for the two lanes on the left from left to right are 1 and 2 respectively, and the horizontal coordinates of the grid matrix for the two lanes on the right from left to right are 4 and 5 respectively. The vertical coordinate of the grid matrix is ​​the safe following distance in front of and behind the vehicle. The vertical coordinate of the grid matrix is ​​the distance s from outside the safe following area in front of the vehicle to a certain distance in front. Similarly, the vertical coordinate of the grid matrix is ​​the distance s from outside the safe following area behind the vehicle to a certain distance behind. The vertical coordinate of the grid matrix is ​​the distance s from the front (rear) of the vehicle to the distance s from the front (rear) of the vehicle. (5) Considering the characteristics of intelligent heavy trucks such as large size, large mass, large inertia, and large turning radius, the grid coverage should be as large as possible to cover the longer braking distance and wider space occupied by the heavy truck.

[0030] Step 2: The traffic participants and obstacles information obtained from multi-source sensors and network information are projected onto the lane to calculate their kinematic characteristics in the Frenet coordinate system, including the Frenet coordinates of the boundary of the perception polygon box, two-dimensional velocity vector, two-dimensional acceleration vector information, and further calculate the grid occupancy rate of participants and obstacles.

[0031] Step 3: Fill the cells of the grid matrix generated in Step 1. Each cell is assigned a multi-dimensional feature, filled using the results from Step 2. Here, a 6-dimensional feature description is used, including but not limited to type, boundary, velocity, acceleration, occupancy rate, and game-theoretic markers. Specifically, when dealing with trailer-mounted heavy trucks, the occupancy rate calculation incorporates the hinge angle as a dynamic parameter, allowing the occupancy rate to reflect the "sweeping" area during turns in real time.

[0032] The grid matrix partitioning described above is not limited to a fixed dimensional structure. Alternatively, the vertical and horizontal grid dimensions (n×m) can be dynamically adjusted based on road type (e.g., highway, urban road), vehicle speed, or real-time computing load. For example, the vertical range can be increased in high-speed scenarios, while the grid resolution can be increased in dense urban scenarios.

[0033] The allocation of grid coordinates is not limited to the vehicle as the center. Alternatively, the grid can be divided with reference to the current lane centerline or the target path.

[0034] The grid shape is not limited to rectangles. Alternatively, Frenet coordinate transformation is performed at curves, with the grid's longitudinal axis coinciding with the road centerline, and the lateral axis always parallel to the normal direction at that point. Longitudinal discretization samples along the arc length of the road centerline. In areas with high curvature (curves), the sampling density is increased (i.e., the longitudinal length of the grid is reduced) to more accurately describe geometric changes; in straight areas, the sampling density can be reduced to improve efficiency. Lateral discretization discretizes along the normal direction of each sampling point on the centerline. The interval of lateral discretization can be fixed or adaptively adjusted according to changes in lane width. The grid naturally conforms to the road shape, more accurately representing the position and movement of traffic participants in curves. In areas with multiple potential paths, such as intersections, instead of relying on a single road centerline, an independent Frenet coordinate system and network are established for each possible driving path (e.g., left turn, straight, right turn). By projecting traffic participants into these different grids, subsequent risk assessments and game-theoretic decisions can be performed in parallel within each possible path grid. The final decision (such as which path to choose) will integrate the assessment results from all grids. In complex intersections or parking lots with unclear topology, the Frenet coordinate system is abandoned in favor of establishing a base coordinate grid centered on the vehicle. The 360-degree area around the vehicle or a specific sector is discretized angularly, and radially discretized from the vehicle's position.

[0035] Dynamic scene representation based on Frenet grids improves the accuracy of environmental modeling, enhances decision robustness, and provides a rich data foundation for subsequent risk quantification and game theory.

[0036] Step 101: Based on the features encoded in the Frenet mesh, quantify lane traffic risk, vehicle-to-vehicle interaction risk, and traffic rule violation risk, and generate a dynamic risk heatmap through spatiotemporal convolution.

[0037] The dynamic risk heatmap is used to characterize the risk level at different times and locations.

[0038] Specifically, when executing step 101, the following steps can be performed: S1010: Calculate the lane flow index and lane speed standard deviation to assess lane-level traffic risk.

[0039] Specifically, the lane flow index is calculated using the following formula: ; In the formula, It is a lane in the grid The number of vehicles inside, It is the first The speed of the car It is the vehicle type coefficient. It is a lane width, It is the free flow velocity. This is an adjustable coefficient. The slope angle is denoted by .

[0040] Preferred, This is the vehicle type coefficient; let's assume it's 1.5 for trucks and 1 for passenger cars. It is a lane The width; This is the free-flow speed, assumed to be 80% of the lane speed limit. This is an adjustable coefficient. The slope angle is significant. When going uphill, heavy trucks move slowly, affecting traffic flow. When going downhill, the braking distance increases dramatically. Therefore, the risks on both uphill and downhill slopes should be amplified. [0,1], the larger the value, the more congested the lane. When the value exceeds a certain threshold, a lane-changing suggestion is triggered.

[0041] The formula for calculating the lane flow index can be adjusted. Alternatively, a more complex traffic flow theory model can be introduced, or a machine learning model based on historical data can be used for predictive congestion assessment. The vehicle type coefficient is not limited to trucks and passenger cars. Alternatively, two-wheeled vehicles, three-wheeled vehicles, etc., can be added and assigned coefficients for non-passenger vehicles. The construction of the lane-changing feasibility index is not limited to the above forms. Alternatively, a model based on collision time and expected minimum clearance can be introduced for comprehensive evaluation.

[0042] Calculate the standard deviation of lane speeds using the following formula: ; In the formula, It is a lane in the grid The number of vehicles inside, It is the first The speed of the car It is a lane The average speed. Risk levels can be categorized into three different levels: low, medium, and high, based on size.

[0043] S1011: Calculate lane-changing feasibility indicators based on the gap and speed difference between vehicles in front and behind, in order to assess the risk of interaction between vehicles.

[0044] Specifically, lane-changing space is assessed based on the following formula to establish lane-changing feasibility indicators, and a comprehensive evaluation of clearance and speed matching is conducted:

[0045] in, , These are the front and rear vehicle clearances, in meters. Let's take the speed difference penalty coefficient as 0.1. Establish two lane-changing decision thresholds, corresponding to prohibiting, being cautious about, and allowing lane changes, respectively.

[0046] By using lane change space assessment metrics, such as clearance and speed matching, to set prohibited / cautious / permitted lane change thresholds, the safety and compliance of lane change decisions can be improved.

[0047] S1012: Based on the traffic rule set, assess potential rule violations to determine the risk of traffic rule violations.

[0048] S1013: Design separable spatiotemporal convolution kernels.

[0049] Among them, the spatial convolution kernel is a Gaussian function whose standard deviation is related to the vehicle speed, and the temporal convolution kernel is a multi-frame sliding window weight vector.

[0050] Specifically, the spatial-temporal joint convolution is as follows:

[0051] Among them, the design can separate convolution kernel groups The spatial convolution kernel is a centrally symmetric Gaussian function:

[0052] With the position of the car Centered on. Standard deviation This is the base value plus a speed compensation item. Among them, Based on the base value, covering the vehicle's braking response delay, the speed compensation term increases the sensitivity range by 5cm for every 1m / s increase in speed. When When it is 0.1, The quantization values ​​are shown in Table 1 below: Table 1: When it is 0.1, quantization value

[0053] Time kernel The structure is The vector, which is the 3-frame sliding window, can be designed as follows: The current frame has the highest weight and is used to reflect the latest environmental state. The previous frame has a weight of 0.3 to smooth out instantaneous noise, and the first two frames are used to capture motion trends.

[0054] Spatiotemporal convolution output risk value:

[0055] S1014: The lane-level traffic risk, vehicle-to-vehicle interaction risk and traffic rule violation risk are weighted and fused, and convolution operation is performed through the spatiotemporal convolution kernel to output the final dynamic risk heat map.

[0056] Specifically, risk values ​​are fused according to the following formula:

[0057] in, It is the risk of lane flow and speed dispersion. , , These are the weighting coefficients; It is a vehicle-to-vehicle interaction risk, which is related to the time to collision (TTC) and the minimum encounter distance (DCPA), and is affected by lane change space. To mitigate the risk of traffic rule violations, key rules to consider include lane markings, right-of-way, and construction zone avoidance.

[0058] In generating dynamic risk heatmaps, the spatiotemporal convolution kernel can alternatively use a Gaussian function to simulate risk diffusion. The number of frames and weights of the temporal convolution kernel can be adjusted, for example, by using exponential smoothing filtering or Kalman filtering to fuse temporal information, instead of a fixed 3-frame sliding window. Risk value, The fusion method is not limited to weighted summation. Alternatively, fuzzy logic, neural networks, etc., can be used to fuse lane flow risks. Interactive risks and rule risks .

[0059] The adaptive dynamic convolution kernel-based risk heatmap generation fully considers lane-level traffic features, inter-vehicle interaction relationships, and spatial topology, and uses spatiotemporal information for risk quantification, enabling more accurate assessment of risks in traffic scenarios.

[0060] Step 102: Construct a non-cooperative game model that includes the vehicle and other traffic participants, use the dynamic risk heat map as a constraint on the strategy space, introduce traffic rule penalty terms into the game payoff function, and obtain the optimal driving strategy of the vehicle by solving the Nash equilibrium.

[0061] Specifically, when executing step 102, the following steps can be performed: S1020: Define the set of game participants and define the strategy space for each game participant.

[0062] The strategy space is composed of a combination of lateral and longitudinal strategies. The lateral strategies include lane keeping, left lane change, and right lane change, while the longitudinal strategies include deceleration, following, and acceleration.

[0063] Specifically, consider non-cooperative game theory among traffic participants. Let the set of participants be denoted as . ,in Indicates the first i Traffic participants.

[0064] S1020: Construct the revenue function.

[0065] The revenue function includes the vehicle's own revenue and the revenue generated through interactions with other participants. The payoff function can preferably be expressed as follows:

[0066] in, For participants i In strategy combination s The following profits, For participants i With participants j Benefits during interaction.

[0067] S1020: Introduce a rule penalty term into the profit function to obtain the modified profit function.

[0068] Specifically, rules to consider include traffic light violations, crossing solid lines, and failure to yield to pedestrians; the corresponding costs could be factored in. They are respectively The revised payoff function obtained after introducing the rule term is as follows:

[0069] in, This is the rule penalty coefficient. The cost of violating rule k is... For indicator functions, when the strategy combination s When rule k is violated, ,otherwise .

[0070] The non-cooperative game model described can be replaced with a partially cooperative game model, which considers the cooperative willingness among traffic participants to seek a collective optimum. The specific driving behavior of the payoff function Ui(s) can be adjusted. Alternatively, different objectives such as safety, comfort, or traffic efficiency can be emphasized based on driving style. The penalty mechanism of the rule term Rrule can also be designed as a continuous function rather than a simple indicator function to handle violations of varying degrees.

[0071] S1020: Based on the dynamic risk heatmap, regions with risk values ​​exceeding the high-risk threshold are marked as infeasible strategy spaces, so as to perform constraint pruning on the strategy space of each participant.

[0072] Specifically, the constraint pruning includes: traversing the strategy space of the participants, removing strategies marked as violating traffic rules by the indicator function in the payoff function; predicting the trajectory of the remaining strategies, and removing the strategy if the predicted trajectory crosses the infeasible strategy space; if the available strategy space of a participant is empty after pruning, adjusting the high-risk threshold and re-screening to ensure that at least one feasible strategy is retained.

[0073] In one specific implementation, a dynamic risk heatmap is used. The space is divided into high / medium / low risk zones, and the strategies of game participants (such as their own vehicle and other vehicles) must avoid high-risk zones. For example, if the risk value of a certain cell is... If the threshold is exceeded, the area is marked as an "infeasible strategy space," and participants cannot choose a path through this area. The risk value is weighted... By integrating lane flow, interaction risks, and rule risks, the complexity of the scenario is directly reflected.

[0074] The strategy space constraints of the participants are denoted as , Let be the constraint function. The participants' policies are horizontal and vertical policy pairs, denoted as . The lateral strategy space consists of {lane keeping, left lane change, right lane change}, and the longitudinal strategy space consists of {deceleration, following, acceleration}. Specifically, in the longitudinal strategy space, deceleration means that after decelerating, the vehicle's relative position in the Frenet grid shifts one grid position backward; acceleration means that after accelerating, the vehicle's relative position in the Frenet grid shifts one grid position forward. Therefore, the finer the Frenet grid, the more efficient the grid division. n The larger the value, the more precise the vehicle's driving actions.

[0075] Participant's strategy space Not limited to the above 3. Discrete action combinations can be alternatively defined as continuous acceleration and steering angle as strategies; the granularity of discrete actions can also be refined, dividing the longitudinal strategy space into rapid acceleration, gradual acceleration, rapid deceleration, gradual deceleration, etc.

[0076] The game strategy space constraint mechanism based on risk heatmaps strongly couples the underlying perceived risk with the high-level game decision-making, transforming risk from a cost term into a direct effect on the generation of the strategy space, thereby improving the security and efficiency of decision-making.

[0077] S1020: Using an iterative strategy, combined with hot-start initialization and constraint pruning, solve for the Nash equilibrium solution of the non-cooperative game model, which is then used as the optimal driving strategy for this vehicle.

[0078] In one specific implementation, an iterative strategy update algorithm is used to solve for the Nash equilibrium, employing a warm-start strategy to initialize the solution using the equilibrium solution from the previous frame. Constraint pruning is implemented to remove obviously illegal strategies in advance during the solution process, thereby accelerating convergence. Specifically, the following sub-steps are included: Obtain the core parameters output from the preceding steps and construct the basic dataset for solving the problem, including: a set of game participants. , Among them are bicycles, Interacting entities include other vehicles and vulnerable road users; the optimized strategy space for each participant. , The strategy feasibility constraint function; the constructed payoff function including rule penalty terms. The generated dynamic risk heatmap satisfies... High-risk spatiotemporal regions; the game equilibrium solution obtained from the previous decision frame. Used for hot start initialization of the current frame.

[0079] The optimal policies of each participant in the previous decision frame are directly assigned as the initial policies for the current frame, i.e.:

[0080] in, Indicates the first i The initial strategy of each participant is completely consistent with its optimal strategy in the previous frame; this warm-start method can make the initial strategy close to the optimal solution of the current scene and reduce the number of iteration convergence steps.

[0081] Iterate through each participant i strategy space For any strategy If there exists at least one class of rules that satisfy the violation indicator function (For violations such as running a red light or changing lanes illegally), the strategy is directly removed, resulting in the pruned strategy space. ; Traversal The remaining strategy is to calculate the driving trajectory corresponding to the strategy through the trajectory prediction model. If the spatiotemporal coordinates of the trajectory at any time are... Falling into a high-risk area If the strategy is discarded, the usable strategy space is obtained. If the available strategy space of a certain participant i If empty, restore the strategy space from the previous pruning round. And adjust the high-risk threshold to Re-screen to ensure that at least one viable strategy is retained.

[0082] An alternating iterative approach is used, with each participant sequentially optimizing their own strategy to maximize the corrected reward until the convergence condition is met. For each participant... Fix the current strategies of all other participants Find its own optimal strategy:

[0083] in, The calculation requires substituting the current strategies of other participants. Combined with the defined interactive benefit items This enables the quantification of benefits under multi-participant interaction.

[0084] Calculate the maximum change in payoffs for all participants before and after the strategy update:

[0085] like If the iteration converges, the current policy set is determined to be... That is, the Nash equilibrium solution in the current scenario; if Then let The strategy for each participant continues to be updated in a loop. If the iteration rounds... If convergence is still not achieved, output the strategy for the current round. As a suboptimal solution, it ensures the real-time performance of the system.

[0086] The converged policy set As the hierarchical game equilibrium solution for the current decision frame, output the optimal strategy for each participant; and then output the equilibrium solution for the current frame. Stored in the cache unit and updated with the historical equilibrium solution for the next decision frame. This is used for hot-start initialization in the next frame solution, forming an iterative optimization closed loop.

[0087] A hierarchical game optimization framework: non-cooperative game + rule constraints, using Nash equilibrium solution strategy, combined with traffic rule penalty terms, to optimize decision results and avoid traffic conflicts caused by local optima of single vehicles, such as intersection deadlock.

[0088] In this embodiment, the continuous and complex traffic environment is discretized into a structured, multi-feature encoded grid representation by constructing a Frenet grid, providing a high-precision and standardized data foundation for subsequent processing and improving the environmental modeling capability and the input quality of the decision-making system. Adaptive spatiotemporal convolution kernels are used to integrate multi-dimensional risk sources such as lane flow, vehicle interaction, and traffic rules to generate a dynamic risk heatmap, enabling refined and real-time quantitative assessment of scenario risks. The risk heatmap is introduced as a hard constraint into a hierarchical non-cooperative game framework, combined with rule-based penalties, so that the solution process for the Nash equilibrium can simultaneously optimize driving efficiency, avoid high-risk areas, and naturally comply with traffic rules. This generates intelligent driving decisions that combine safety, compliance, efficiency, and strong interactive adaptability in complex dynamic scenarios.

[0089] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0090] like Figure 2 As shown, the following are embodiments of the heavy-duty truck decision-making system based on Frenet grid and hierarchical dynamic game provided by this disclosure. These embodiments belong to the same inventive concept as the heavy-duty truck decision-making methods based on Frenet grid and hierarchical dynamic game in the above embodiments. For details not described in detail in the embodiments of the heavy-duty truck decision-making system based on Frenet grid and hierarchical dynamic game, please refer to the embodiments of the heavy-duty truck decision-making methods based on Frenet grid and hierarchical dynamic game described above.

[0091] A heavy-duty truck decision-making system based on Frenet grid and hierarchical dynamic game theory includes: The Frenet mesh building module is used to discretize the road space around the vehicle into a Frenet mesh, wherein the Frenet mesh includes multiple cells and encodes multidimensional features for each cell; The risk quantification module is used to quantify lane traffic risk, vehicle-to-vehicle interaction risk, and traffic rule violation risk based on the features encoded in the Frenet grid, and to generate a dynamic risk heat map through spatiotemporal convolution, wherein the dynamic risk heat map is used to characterize the risk level at different spatiotemporal locations. The game optimization module is used to construct a non-cooperative game model that includes the vehicle itself and other traffic participants. The dynamic risk heat map is used as a constraint on the strategy space, and traffic rule penalty terms are introduced into the game payoff function. The optimal driving strategy of the vehicle is obtained by solving the Nash equilibrium.

[0092] Figure 3 This is a schematic diagram of the hardware structure of an electronic device that implements various embodiments of the present invention.

[0093] The heavy-duty truck decision-making method based on Frenet grid and hierarchical dynamic game theory provided in this application can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiments of this invention does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, the electronic device includes, but is not limited to, laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0094] Electronic devices may include processors, external memory interfaces, internal memory, universal serial bus (USB) interfaces, charging management modules, power management modules, batteries, wireless communication modules, audio modules, speakers, microphones, sensor modules, buttons, cameras, displays, and SIM card interfaces, etc.

[0095] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the electronic device. In other embodiments of this application, the electronic device may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0096] A processor may include one or more processing units, such as: a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0097] The processor can serve as the nerve center and command center of an electronic device. The controller can generate operation control signals based on the instruction opcode and timing signals to control the fetching and execution of instructions.

[0098] The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0099] An external storage interface (ESI) can be used to connect external memory cards, such as microSD cards, to expand the storage capacity of electronic devices. The external memory card communicates with the processor through the ESI to perform data storage functions, such as saving music and video files on the external memory card.

[0100] Internal memory can be used to store computer executable program code, which includes instructions. The processor executes various functional applications and data processing of electronic devices by running the instructions stored in internal memory. Internal memory can include a program storage area and a data storage area. Internal memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, universal flash storage (UFS), etc.

[0101] Wireless communication functionality in electronic devices can be achieved through antennas, wireless communication modules, modem processors, and baseband processors.

[0102] Wireless communication modules can provide solutions for wireless communication applications in electronic devices, including wireless local area networks (WLANs) (such as wireless fidelity (Wi-Fi) networks), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), and infrared (IR) technologies.

[0103] Electronic devices can implement audio functions through audio modules, speakers, receivers, microphones, headphone jacks, and application processors.

[0104] Electronic devices can achieve shooting functions through ISPs, cameras, video codecs, GPUs, displays, and application processors.

[0105] Electronic devices can achieve display functions through GPUs, displays, and application processors.

[0106] A GPU is a microprocessor for image processing, connected to the display screen and application processor. GPUs are used to perform mathematical and geometric calculations for graphics rendering. A processor may include one or more GPUs, which execute program instructions to generate or modify display information.

[0107] A display screen is used to display images, videos, etc. A display screen includes a display panel.

[0108] The storage medium provided in this application stores a program product capable of implementing a heavy truck decision-making method based on Frenet grid and hierarchical dynamic game.

[0109] The heavy-duty truck decision-making method based on Frenet grids and hierarchical dynamic game theory includes: discretizing the road space around the vehicle into a Frenet grid, wherein the Frenet grid comprises multiple cells, and each cell is encoded with multidimensional features; based on the features encoded in the Frenet grid, quantifying lane traffic risk, inter-vehicle interaction risk, and traffic rule violation risk, and generating a dynamic risk heatmap through spatiotemporal convolution, wherein the dynamic risk heatmap is used to characterize the risk level at different spatiotemporal locations; constructing a non-cooperative game model including the vehicle and other traffic participants, using the dynamic risk heatmap as a constraint condition in the strategy space, and introducing a traffic rule penalty term into the game payoff function, and obtaining the optimal driving strategy of the vehicle by solving the Nash equilibrium.

[0110] In some possible implementations, the subject matter of this disclosure, namely, "A method and system for heavy-duty truck decision-making based on Frenet grids and hierarchical dynamic games," can be implemented as a program product comprising program code that, when run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0111] The storage medium disclosed herein may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0112] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A heavy-duty truck decision-making method based on Frenet grid and hierarchical dynamic game, characterized in that, include: The road space surrounding the vehicle is discretized into a Frenet grid, wherein the Frenet grid comprises multiple cells, and each cell is encoded with multidimensional features; Based on the features encoded in the Frenet mesh, lane traffic risk, vehicle-to-vehicle interaction risk, and traffic rule violation risk are quantified, and a dynamic risk heatmap is generated through spatiotemporal convolution. The dynamic risk heatmap is used to characterize the risk level at different spatiotemporal locations. A non-cooperative game model involving the vehicle and other traffic participants is constructed. The dynamic risk heatmap is used as a constraint on the strategy space, and traffic rule penalty terms are introduced into the game payoff function. The optimal driving strategy of the vehicle is obtained by solving the Nash equilibrium.

2. The heavy truck decision-making method based on Frenet grid and hierarchical dynamic game as described in claim 1, characterized in that, The road space surrounding the vehicle is discretized into a Frenet mesh, including: Centered on the vehicle, the longitudinal and transverse road space is discretized into a grid matrix, where the longitudinal direction represents the road direction and the transverse direction represents the lane direction; The information of traffic participants and obstacles is projected onto the Frenet grid matrix to obtain the kinematic characteristics of traffic participants and obstacles under the Frenet grid matrix, and the grid occupancy rate of traffic participants and obstacles is calculated. Based on the grid occupancy rates of the traffic participants and obstacles, a multidimensional feature vector is filled into each cell of the Frenet grid matrix. The multidimensional feature vector includes at least: object type, boundary, velocity vector, acceleration vector, occupancy rate, and game marker.

3. The heavy truck decision-making method based on Frenet grid and hierarchical dynamic game as described in claim 1, characterized in that, Dynamic risk heatmaps are generated through spatiotemporal convolution, including: Calculate the lane flow index and lane speed standard deviation to assess lane-level traffic risk; Based on the gap and speed difference between vehicles in front and behind, lane-changing feasibility indicators are calculated to assess the risk of interaction between vehicles. Based on a set of traffic rules, potential rule violations are assessed to determine the risk of traffic rule violations. Design a separable spatiotemporal convolution kernel, wherein the spatial convolution kernel is a Gaussian function whose standard deviation is related to the vehicle speed, and the temporal convolution kernel is a multi-frame sliding window weight vector; The lane-level traffic risks, vehicle-to-vehicle interaction risks, and traffic rule violation risks are weighted and fused together, and then convolutional operations are performed using the spatiotemporal convolution kernel to output the final dynamic risk heatmap.

4. The heavy truck decision-making method based on Frenet grid and hierarchical dynamic game as described in claim 3, characterized in that, The lane flow index is calculated using the following formula: ； In the formula, It is a lane in the grid The number of vehicles inside, It is the first The speed of the car It is the vehicle type coefficient. It is a lane width, It is the free flow velocity. This is an adjustable coefficient. The slope angle is denoted by .

5. The heavy truck decision-making method based on Frenet grid and hierarchical dynamic game as described in claim 3, characterized in that, Calculate the standard deviation of lane speeds using the following formula: ； In the formula, It is a lane in the grid The number of vehicles inside, It is the first The speed of the car It is a lane The average speed.

6. The heavy truck decision-making method based on Frenet grid and hierarchical dynamic game as described in claim 1, characterized in that, A non-cooperative game model is constructed, incorporating the vehicle itself and other traffic participants. The dynamic risk heatmap is used as a constraint on the strategy space, and traffic rule penalties are introduced into the game payoff function. The optimal driving strategy for the vehicle is obtained by solving for the Nash equilibrium, including: Define a set of game participants and define a strategy space for each game participant. The strategy space consists of a combination of lateral and vertical strategies. Lateral strategies include lane keeping, left lane change, and right lane change. Vertical strategies include deceleration, following, and acceleration. Construct a revenue function, wherein the revenue function includes the vehicle's own revenue and the revenue from interactions with other participants; By introducing a rule penalty term into the profit function, a modified profit function is obtained. Based on the dynamic risk heatmap, areas with risk values ​​exceeding the high-risk threshold are marked as infeasible strategy spaces, so as to perform constraint pruning on the strategy space of each participant. An iterative strategy is adopted, combining hot-start initialization and constraint pruning, to solve the Nash equilibrium solution of the non-cooperative game model, which is then used as the optimal driving strategy for this vehicle.

7. The heavy truck decision-making method based on Frenet grid and hierarchical dynamic game as described in claim 6, characterized in that, The constrained pruning includes: Traverse the participants' strategy space, eliminate strategies marked as violating traffic rules by the indicator function in the payoff function; and predict the trajectory of the remaining strategies. If the predicted trajectory crosses the infeasible strategy space, then eliminate the strategy. If a participant's available strategy space is empty after pruning, the high-risk threshold is adjusted and the selection is re-screened to ensure that at least one feasible strategy is retained.

8. A heavy-duty truck decision-making system based on Frenet grid and hierarchical dynamic game, characterized in that, include: The Frenet mesh building module is used to discretize the road space around the vehicle into a Frenet mesh, wherein the Frenet mesh includes multiple cells and encodes multidimensional features for each cell; The risk quantification module is used to quantify lane traffic risk, vehicle-to-vehicle interaction risk, and traffic rule violation risk based on the features encoded in the Frenet grid, and to generate a dynamic risk heat map through spatiotemporal convolution, wherein the dynamic risk heat map is used to characterize the risk level at different spatiotemporal locations. The game optimization module is used to construct a non-cooperative game model that includes the vehicle itself and other traffic participants. The dynamic risk heat map is used as a constraint on the strategy space, and traffic rule penalty terms are introduced into the game payoff function. The optimal driving strategy of the vehicle is obtained by solving the Nash equilibrium.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the heavy truck decision-making method based on Frenet grid and hierarchical dynamic game as described in any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the heavy truck decision-making method based on Frenet grid and hierarchical dynamic game as described in any one of claims 1 to 7.

Images