Intelligent automobile dynamic scene planning method and system based on time-sharing operation

Abstract

The invention provides an intelligent automobile dynamic scene planning method and system based on time-sharing operation, and aims to solve the problems of calculation real-time performance and result effectiveness of an existing global path planning method. The method comprises four architectures of information processing, short-term prediction, time-sharing operation and path planning. Compared with an existing global planning method, the method has the advantages that dynamic changes of a scene are recognized through time-sharing operation, operation triggering of A * is set to be two parts of local dynamic correction and overall increment updating, and the problems that a traditional algorithm is poor in single-time operation precision and large in continuous operation resource consumption are effectively solved. Meanwhile, a short-term prediction model is constructed based on a nonlinear vehicle kinetic equation, and a prediction result is used for scene future grid construction, so that the effectiveness of an operation result can be effectively improved; through scene time-sharing stage division, the method can flexibly adapt to continuously changing road environments, and ensures that the automatic driving vehicle always follows the optimal and safe driving path.

Description

Technical Field

[0001] The present invention relates to a method for dynamic scene planning of an intelligent vehicle, and in particular to a method and system for dynamic scene planning of an intelligent vehicle based on time-sharing operation. Background Art

[0002] Smart car technology is an important strategic high ground in the future travel field. It is committed to providing humans with unprecedented convenient, safe and comfortable mobile experience. In this technical framework, path planning plays a vital role, that is, by integrating the current status information of the vehicle, the coordinates of the destination and the real-time environmental data, to accurately calculate a driving trajectory that meets the driving constraints and is optimized as much as possible. Path planning methods are usually divided into two levels: global planning and local planning. Global planning is to generate a rough path from the starting point to the end point within the entire map according to the road topology and traffic rules. Local planning is based on global planning, based on the dynamic model of the car and the position of surrounding obstacles, to generate a more refined and smooth path.

[0003] The global planning method plays a decisive role in the intelligent vehicle decision-making system. It lays the foundation for the overall driving direction and action strategy of the vehicle, and provides key guiding principles and boundary conditions for subsequent local planning. However, the current global planning algorithms have exposed some limitations when facing dynamically changing driving scenarios, which are mainly reflected in the high computational complexity and long operation time. Therefore, they can often only provide a relatively static path guidance. Since global planning needs to deal with a huge search space and complex optimization problems, the huge computational load generated by this delays the ability to respond to rapid changes in the environment, thereby weakening the real-time nature of the path planning results. Therefore, it is urgent to develop a new global planning method that can efficiently cope with dynamic changes in the environment. Through powerful real-time computing performance and scene perception adaptability, it can quickly identify and adapt to the changing road environment, and then efficiently calculate the most suitable driving path, thereby effectively supporting intelligent vehicles to make more accurate and timely decisions and control actions. Summary of the invention

[0004] In order to solve the problems of computational real-time performance and result validity in the above-mentioned existing global path planning methods, the present invention provides a smart car dynamic scene planning method based on time-sharing operation, comprising the following steps:

[0005] Step 1: Scene information processing and key element extraction: Process the scene information input by the sensor, perform road grid conversion, build a basic grid, and extract key elements in the scene;

[0006] Step 2. Short-term motion prediction judgment and output: Based on basic kinematics, a prediction model is constructed. Given the current vehicle state parameters, the future driving instructions are calculated based on the historical states. Integral operations are performed according to the kinematic prediction model equation to obtain the vehicle state at future times, and the stage parameters of the moving object within a preset future time period are output to fill the scene grid.

[0007] Step 3. Scene time-sharing stage division: Using the prediction data and scene cycle to identify the time-sharing characteristics and perform time-sharing operation stage division, including:

[0008] 31) Scene cycle calculation: Set to execute the global planning algorithm once within a fixed cycle; set a fixed scene update cycle T, and save the current global trigger time Tc each time the operation is completed; if the actual running time t = n×T + Tc, then proceed to the next round of operation process and perform an overall incremental update of the scene state based on the latest environmental state.

[0009] 32) After the A* algorithm calculates the optimal driving path, if other vehicles exhibit abnormal behaviors and interfere during this process, judge the next decision according to the interference degree. The method is as follows:

[0010] Perform dilation processing on the trajectory calculated by A*, set the original trajectory width Lex, and delimit the driving area of the host vehicle.

[0011] Judge the current and future motion conditions of the moving objects in the scene. If the motion area invades the driving area of the host vehicle, take the point closest to the trajectory center as the intrusion point, calculate the distance Lin from the intrusion point to the trajectory center. If Lin > Lex / 2, it means excessive intrusion, and directly perform an overall incremental update; if Lin ≤ Lex / 2, calculate the remaining distance outside the trajectory, extend it outward by twice the vehicle length until an obstacle is encountered, and record the outer remaining distance Lout. If Lout < Lin, it means the outer distance is insufficient, and directly perform an overall incremental update. If Lout ≥ Lin, perform local dynamic correction.

[0012] 33) Save the scene time-sharing operation results: According to the above analysis and decision, execute the corresponding path correction measures and save the optimized time-sharing operation results as the basis for the next moment's decision.

[0013] Step 4. Initialization of the A* algorithm architecture: Initialize the nodes of the A* algorithm and assign corresponding attributes, set the corresponding cost function and heuristic function for each node, set the iteration depth, and define the bidirectional optimization heuristic function.

[0014] Step 5. Bidirectional iterative loop search for the path: Search for the path bidirectionally, gradually iterate and deepen the loop process, and merge the bidirectional paths to achieve rapid planning.

[0015] Step 6. Dynamic update of scene and path output: When the calculation status of step 3 is local dynamic correction, the local dynamic correction strategy is adopted. When the calculation status of step 3 is overall incremental update, the overall incremental update strategy is adopted. The dynamic changes of the scene are responded to through local dynamic correction or overall incremental update, and the final path is output.

[0016] Preferably, step 1 of scene information processing and key element extraction specifically includes:

[0017] 11) Raw data acquisition and preprocessing: Real-time environmental data is obtained from vehicle-mounted sensors as raw data, the raw data is denoised, and the coordinate system is converted to the same reference coordinate system.

[0018] 12) Determine the resolution of the grid map based on the actual application scenario: The resolution of the grid map should be as large as possible, but should be less than half the width of the vehicle, less than half the minimum turning radius of the vehicle, and less than the maximum distance the vehicle can travel in a perception and control cycle;

[0019] 13) Extract key scene elements: Extract the position and size information of traffic participants and obstacles related to the movement of the main vehicle in the scene for updating the grid map. According to the sensor data, each grid is assigned an attribute value to represent the characteristics of the grid.

[0020] Features include the road conditions at the location, whether there is a car at the location, and auxiliary information of the location.

[0021] Preferably, step 2, short-term motion prediction judgment and output specifically includes:

[0022] 21) Construct a basic kinematic prediction model: A nonlinear kinematic model is used to describe vehicle motion. The basic structure of the model is as follows: ① The vehicle consists of two wheels, and the direction is controlled by the front wheel steering angle δ; ② The position of the vehicle's center of mass is represented by the coordinates (x, y), and the heading angle is represented by θ; the dynamic equation is as follows:

[0023] Longitudinal dynamics include:

[0024]

[0025]

[0026] Where v represents the vehicle speed, i.e., the speed of the center of mass along the vehicle's forward direction, and a represents the vehicle acceleration; lateral dynamics include:

[0027]

[0028]

[0029] Among them, ω represents the vehicle yaw rate, β represents the steering slip angle, L is the wheelbase, and r is the turning radius;

[0030] Using the relationship in the Frenet coordinate system, we can deduce:

[0031]

[0032] The longitudinal dynamics and lateral dynamics equations are combined to form a complete kinematic prediction model, which is used to calculate the vehicle state (x, y, θ, v, ω) at future moments.

[0033] 22) Kinematic short-term prediction output: Given the current vehicle state parameters (x(0), y(0), θ(0), v(0), ω(0)), the future driving instructions are obtained based on the historical state calculations. The vehicle state at the future time t is obtained by performing integral calculations based on the kinematic prediction model equations, including position, speed, and heading angle. The calculation formula is as follows:

[0034] x(t)=x(0)+∫[v]dt

[0035] y(t)=y(0)+∫[v×sin(θ+β)]dt

[0036] θ(t)=θ(0)+∫[ω]dt

[0037] v(t)=v(0)+∫[a]dt

[0038]

[0039] Thus, the vehicle trajectory data within the future preset time period is output, including the position (x, y) and posture (θ) information at each time node.

[0040] 23) Scene grid mapping: Fill the output vehicle trajectory data into the rasterized map of step 1.

[0041] Preferably, step 4, initializing the A* algorithm architecture specifically includes:

[0042] 41) Initialize the start node (S) and the target node (G), and assign corresponding attributes, including coordinates and states; set the corresponding cost function g(n) and heuristic function h(n) for each node.

[0043] 42) Build data structure: Create two open lists (Open List), corresponding to the forward search from the starting point (S) to the end point (G) and the reverse search from the end point (G) to the starting point (S); at the same time, create two closed lists (Closed List) to record the information of the nodes that have been visited and processed, to ensure that repeated calculations and invalid traversals are avoided during the search process.

[0044] 43) Set the iteration depth and bidirectional optimization heuristic function: Set the initial iteration depth, and increase this limit by a certain value each time the iteration deepens; bidirectional optimization requires symmetry to ensure that the search process can be effectively guided in both directions;

[0045] Furthermore, Manhattan distance is adopted as the heuristic function to ensure the consistency of bidirectional search.

[0046] For two points (x1, y1) and (x2, y2) on a two-dimensional plane, the Manhattan distance is defined as:

[0047] h manhattan =|x1-x2|+|y1-y2|

[0048] Preferably, the bidirectional iterative loop search of the path in step 5 specifically includes:

[0049] 51) Bidirectional search execution: forward A search is performed from the starting point, and reverse A search is performed from the target node at the same time; in each round of iteration, the node with the highest priority (i.e., the node with the smallest F value) is selected for expansion according to F(n)=g(n)+h(n).

[0050] 52) Gradually iterate and deepen the loop process: When a round of search cannot find a solution or reaches the current iteration depth limit, increase the iteration depth limit and continue to the next round of search; during each iteration, the open list and closed list must be maintained, and the cost information of the node must be updated.

[0051] 53) Path intersection detection and merging: When the forward search front and the reverse search front meet, it means that a possible shortest path has been found; trace back to their respective start / target nodes in the forward and reverse directions to form two half paths, which are then spliced ​​together to obtain the complete optimal path.

[0052] Preferably, step 6, scene dynamic update and path output specifically includes:

[0053] 61) When the calculation status of step three is local dynamic correction, the local dynamic correction strategy is adopted, including: a. calculating the change of the cost function of the affected section, and adjusting the G value and F value of the corresponding node according to the new environmental information; b. recalculating the shortest path of the affected part based on the existing path, and performing local optimization; c. updating the open list and closed list to reflect the latest path status.

[0054] 62) When the calculation state in step three is an overall incremental update, an overall incremental update strategy is adopted, including: a. resetting the starting point and target point according to the new environmental information; b. resetting the search process and re-executing the bidirectional A* search according to the new iteration depth limit; c. in the new search process, if it is not a global update caused by the scene periodic time sharing, then using known valid information to accelerate the search convergence.

[0055] 63) Output the final path.

[0056] The present invention also provides a smart car dynamic scene planning system based on time-sharing operation, comprising an information processing module, a short-term prediction module, a time-sharing operation module and a path planning module; the information processing module processes the scene information input by the sensor, performs road grid conversion, constructs a basic grid, and extracts key elements in the scene, specifically performing the operation of the above step one; the short-term prediction module constructs a prediction model based on basic kinematics, predicts the phased parameters of the moving subject, fills the scene grid, and specifically performs the operation of the above step two; the time-sharing operation module uses the prediction data and the scene cycle to identify the time-sharing characteristics, divides the scene into stages, and specifically performs the operation of the above step three; the path planning module first initializes the nodes and iteration depth of the A* algorithm, defines a bidirectional optimization heuristic function, and then bidirectionally searches for paths, gradually iterates and deepens the cycle process, merges bidirectional paths to achieve rapid planning, and finally responds to dynamic changes in the scene through local dynamic correction and overall incremental update, outputs the final path result, and specifically performs the operations of the above steps four to six; and finally enables the smart car to complete an efficient and safe driving process.

[0057] Beneficial effects of the present invention:

[0058] The present invention provides a method and system for dynamic scene planning of intelligent vehicles based on time-sharing operation, which includes a four-part architecture of an information processing module, a short-term prediction module, a time-sharing operation module and a path planning module. Compared with the existing global planning method, the present invention identifies the dynamic changes of the scene through time-sharing operation, and sets the operation trigger of A* to two parts: local dynamic correction and overall incremental update, which effectively solves the problems of poor single-time operation accuracy and large continuous operation resource consumption of traditional algorithms. At the same time, the present invention constructs a short-term prediction model based on nonlinear vehicle dynamics equations, and uses the prediction results for future grid construction of the scene, which can effectively improve the effectiveness of the operation results; the present invention can flexibly adapt to the ever-changing road environment by dividing the scene into time-sharing stages, ensuring that the autonomous driving vehicle always follows the optimal and safe driving path. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0060] Figure 1 It is a schematic diagram of the process of the present invention;

[0061] Figure 2 This is a schematic diagram of the dynamic interference effect of the scene of the present invention;

[0062] Figure 3 It is a schematic diagram of the time-sharing operation flow of the present invention;

[0063] Figure 4 Schematic diagram of the system architecture of the present invention. DETAILED DESCRIPTION

[0064] Embodiment 1,

[0065] like Figure 1 As shown, this embodiment provides a method for dynamic scene planning of an intelligent vehicle based on time-sharing operation, comprising the following steps:

[0066] Step 1: Scene information processing and key element extraction.

[0067] Process the scene information input by the sensor, perform road grid conversion, build the basic grid, and extract the key elements in the scene, including:

[0068] 11) Raw data acquisition and preprocessing: Real-time environmental data is obtained from vehicle-mounted sensors (such as lidar, camera, millimeter-wave radar, ultrasonic sensor, GPS, etc.) as raw data, and the raw data is denoised and converted to the same reference coordinate system.

[0069] 12) Determine the resolution of the grid map according to the actual application scenario: In high-speed scenarios, the control process can be divided into two parts: longitudinal speed control and lateral lane change control. In order to clearly represent the shape and position of the vehicle, the resolution of the grid map should be less than half of the width of the vehicle; in order to ensure the vehicle's motion characteristics and constraints, the resolution of the grid map should be less than half of the vehicle's minimum turning radius; in order to adapt to the vehicle's speed and acceleration, the resolution of the grid map should be less than the maximum distance the vehicle may travel in a perception and control cycle; in order to save data volume and computing resources, the resolution of the grid map should be as large as possible, but not exceeding the above limits.

[0070] 13) Extract key scene elements: Extract the position and size information of traffic participants and obstacles related to the main vehicle movement in the scene for updating the grid map. According to the sensor data, attribute values ​​are assigned to each grid to represent the characteristics of the grid. This embodiment is represented by an array containing 4 elements, such as [0,1,1,1], where the first number indicates the road condition at the location (0 indicates a normal traffic section, 1 indicates that the location is prohibited from entering), the second number indicates whether there is a car at the location (0 indicates no car, 1 indicates a car), the third number indicates the ancillary information of the location (if there is a car at the location, it indicates whether the vehicle is the main vehicle or other traffic vehicles), and the fourth number indicates further ancillary information (indicating whether the area is related to vehicle movement).

[0071] Step 2: Short-term motion prediction, judgment and output.

[0072] Based on basic kinematics, a prediction model is built to predict the phased parameters of the moving subject and fill the scene grid, including:

[0073] 21) Constructing a basic kinematic prediction model: In order to balance computational efficiency and application effects in complex scenarios, this embodiment selects a nonlinear kinematic model to describe vehicle motion; a bicycle model is selected, and the basic structure of the model is as follows: ① The vehicle consists of two wheels, and the direction is controlled by the front wheel steering angle δ; ② The position of the vehicle's center of mass is represented by the coordinates (x, y), and the heading angle is represented by θ; the dynamic equation is as follows:

[0074] Longitudinal dynamics include:

[0075]

[0076]

[0077] Where v represents the vehicle speed, i.e., the speed of the center of mass along the vehicle's forward direction, and a represents the vehicle acceleration; lateral dynamics include:

[0078]

[0079]

[0080] Among them, ω represents the vehicle yaw rate, β represents the steering slip angle, L is the wheelbase, and r is the turning radius;

[0081] Using the relationship in the Frenet coordinate system, we can deduce:

[0082]

[0083] The longitudinal dynamics and lateral dynamics equations are combined to form a complete kinematic prediction model, which is used to calculate the vehicle state (x, y, θ, v, ω) at future moments.

[0084] 22) Kinematic short-term prediction output: Given the current vehicle state parameters (x(0), y(0), θ(0), v(0), ω(0)), the future driving instructions are obtained based on the historical state calculations. The vehicle state at the future time t is obtained by performing integral calculations based on the kinematic prediction model equations, including position, speed, and heading angle. The calculation formula is as follows:

[0085] x(t)=x(0)+∫[v]dt

[0086] y(t)=y(0)+∫[v×sin(θ+β)]dt

[0087] θ(t)=θ(0)+∫[ω]dt

[0088] v(t)=v(0)+∫[a]dt

[0089]

[0090] Thus, the vehicle trajectory data within the future preset time period is output, including the position (x, y) and posture (θ) information at each time node.

[0091] 23) Scene grid mapping: Fill the output vehicle trajectory data into the rasterized map of step 1.

[0092] Step 3: Divide the scene into time stages.

[0093] like Figure 2 , 3 As shown in the figure, the time-sharing characteristics are identified by using the prediction data and the scenario cycle, and the time-sharing operation stages are divided, including:

[0094] 31) Scene cycle calculation: In order to reduce the amount of calculation, improve the calculation efficiency and maintain the real-time performance of path planning, the global planning algorithm is only executed once within a fixed period; a fixed scene update period T is set, and the current global trigger time Tc is saved after each calculation is completed; then the actual running time t is monitored. If t = n × T + Tc, where n is a natural number, the next round of calculation process is carried out, and the overall scene status is incrementally updated based on the latest environmental status.

[0095] 32) After the A* algorithm calculates the best driving path, if other vehicles have abnormal behavior interference during the process, it will affect the feasibility of the path. Therefore, the next decision needs to be determined based on the degree of interference. The specific method is as follows:

[0096] Trajectory Buffer Construction: Inflate the trajectory calculated by the A* algorithm, and set the original trajectory width Lex to 1.5 times the width of the host vehicle, and delimit the driving area of the host vehicle to ensure sufficient obstacle avoidance space.

[0097] Dynamic Object Intrusion Detection: Monitor the current and future movement trends of other moving objects in the scene. Once their movement areas overlap with the safe driving area of the host vehicle, it is determined as an intrusion event.

[0098] Determine the Severity of Intrusion: For the moving object that has intruded, find the contact point closest to the center of the host vehicle's driving trajectory, and calculate the distance Lin from this intrusion point to the trajectory center. If Lin > Lex / 2, it indicates that the intrusion situation is serious, and directly trigger the overall incremental update process to re-plan the path.

[0099] Available Space Evaluation: If Lin ≤ Lex / 2, then further evaluate the available space outside the host vehicle's trajectory. Expand the distance twice the vehicle length outward from the original trajectory until an obstacle is encountered, and record the available distance Lout at this time; if Lout < Lin, it means that the host vehicle does not have enough space to detour, and the overall incremental update also needs to be performed immediately; on the contrary, if Lout ≥ Lin, adjust the current path of the host vehicle through the local dynamic correction method to avoid the conflict area.

[0100] 33) Save the Results of Scene Time-Sharing Computation: According to the above analysis and decision-making, execute the corresponding path correction measures, and save the time-sharing computation results after optimization processing as the basis for decision-making at the next moment. In this way, the system can flexibly adapt to the changing road environment and ensure that the autonomous vehicle always follows the optimal and safe driving path.

[0101] Step Four: Initialize the A* Algorithm Architecture.

[0102] Initialize the nodes and iteration depth of the A* algorithm, and define the bidirectional optimization heuristic function, specifically including:

[0103] 41) Initialize the starting node (S) and the target node (G), and assign corresponding attributes, including coordinates and status; set the corresponding cost function g(n) (the actual cost from the starting point to the current node) and heuristic function h(n) (the estimated cost from the current node to the target node) for each node.

[0104] 42) Construct the data structure: Create two open lists, corresponding to the forward search from the starting point (S) to the end point (G) and the reverse search from the end point (G) to the starting point (S) respectively; at the same time, create two closed lists to record the information of the nodes that have been visited and processed, ensuring to avoid repeated calculations and invalid traversals during the search process.

[0105] 43) Set the iteration depth and bidirectional optimization heuristic function: Set the initial iteration depth to 2, and increase this limit by a certain value each time the iteration is deepened; bidirectional optimization requires symmetry to ensure that the search process can be effectively guided in both directions. Therefore, this embodiment uses Manhattan distance as a heuristic function to ensure the consistency of bidirectional search, which means that whether it is forward or reverse search, the same Manhattan distance formula will be used to estimate the distance from the current node to the target node, thereby achieving efficient guidance and synchronous convergence of bidirectional search. In this way, while ensuring the search accuracy, the search speed can be significantly improved, and it can adapt to the needs of dynamically changing scenarios.

[0106] Step 5: Bidirectional iterative loop search of the path.

[0107] Bidirectional search path, gradually iterate and deepen the cycle process, merge bidirectional paths to achieve rapid planning, specifically including:

[0108] 51) Bidirectional search execution: forward A search is performed from the starting point, and reverse A search is performed from the target node at the same time; in each round of iteration, the node with the highest priority (i.e., the node with the smallest F value) is selected for expansion according to F(n)=g(n)+h(n).

[0109] 52) Gradually iterate and deepen the loop process: When a round of search cannot find a solution or reaches the current iteration depth limit, increase the iteration depth limit and continue to the next round of search; during each iteration, the open list and closed list must be maintained, and the cost information of the node must be updated.

[0110] 53) Path intersection detection and merging: When the forward search front and the reverse search front meet (theoretically at an intermediate node), it means that a possible shortest path has been found. Go back to their respective start / target nodes in the forward and reverse directions to form two half-paths, which are then spliced ​​together to obtain the complete optimal path.

[0111] Step 6: Dynamic scene update and path output.

[0112] Through local dynamic correction and overall incremental update, the dynamic changes of the scene can be dealt with and the final path output can be performed, including:

[0113] 61) When the calculation status of step three is local dynamic correction, the local dynamic correction strategy is adopted, including: a. calculating the change of the cost function of the affected section, and adjusting the G value and F value of the corresponding node according to the new environmental information; b. recalculating the shortest path of the affected part based on the existing path, and performing local optimization; c. updating the open list and closed list to reflect the latest path status.

[0114] 62) When the calculation status of step three is the overall incremental update, the overall incremental update strategy is adopted, including: a. Resetting the starting point (current position) and target point according to the new environmental information (may need to be adjusted to the new optimal target); b. Resetting the search process and re-executing the bidirectional A* search according to the new iteration depth limit; c. In the new search process, if it is not a global update caused by the scene periodic time sharing, use the known valid information (such as the low-cost path fragments in the previous search results) to accelerate the search convergence.

[0115] 63) Output the final path.

[0116] Embodiment 2,

[0117] like Figure 4 As shown, the present embodiment provides a dynamic scene planning system for an intelligent vehicle based on time-sharing operation, including an information processing module, a short-term prediction module, a time-sharing operation module and a path planning module; the information processing module processes the scene information input by the sensor, performs road grid conversion, constructs a basic grid, and extracts key elements in the scene, specifically performing the operation of step 1 of the above embodiment 1; the short-term prediction module constructs a prediction model based on basic kinematics, predicts the phased parameters of the moving subject, fills the scene grid, and specifically performs the operation of step 2 of the above embodiment 1; the time-sharing operation module uses the prediction data and the scene cycle to identify the time-sharing characteristics, divides the scene into stages, and specifically performs the operation of step 3 of the above embodiment 1; the path planning module first initializes the nodes and iteration depth of the A* algorithm, defines a bidirectional optimization heuristic function, and then bidirectionally searches for paths, gradually iterates and deepens the cycle process, merges bidirectional paths to achieve rapid planning, and finally responds to dynamic changes in the scene through local dynamic correction and overall incremental update, and outputs the final path result, specifically performing the operations of steps 4 to 6 of the above embodiment 1; and finally enables the intelligent vehicle to complete an efficient and safe driving process.

Claims

1. A method for dynamic scene planning of intelligent vehicles based on time-sharing computing, characterized in that: It includes the following steps: Step 1, Scene information processing and key element extraction: Process the scene information input by the sensor, perform road grid conversion, construct a basic grid, and extract the key elements in the scene; Step 2, Short-term motion prediction judgment and output: Based on basic kinematics, construct a prediction model. Given the current vehicle state parameters, calculate the future driving instructions according to the historical state, perform integral operations according to the kinematic prediction model equation to obtain the vehicle state at future times, output the stage parameters of the moving object within a preset future time period, and fill the scene grid; Step 3, Scene time-sharing stage division: Use the prediction data and scene cycle to identify time-sharing characteristics and perform time-sharing operation stage division, including: 31) Scene cycle calculation: Set to execute the global planning algorithm once within a fixed cycle; Set a fixed scene update cycle T, and save the current global trigger time Tc every time the operation is completed; If the actual running time t = n×T + Tc, then perform the next round of operation process and perform an overall incremental update of the scene state based on the latest environmental state; 32) After the A* algorithm calculates the optimal driving path, if other vehicles have abnormal behavior interference during this process, judge the next decision according to the interference degree. The method is as follows: Perform dilation processing on the trajectory calculated by A*, set the original trajectory width Lex, and delimit the driving area of the host vehicle; Judge the current and future motion conditions of the moving objects in the scene. If the motion area invades the driving area of the host vehicle, take the point closest to the trajectory center as the intrusion point, calculate the distance Lin from the intrusion point to the trajectory center. If Lin > Lex / 2, it means excessive intrusion, and directly perform an overall incremental update; If Lin ≤ Lex / 2, calculate the remaining distance outside the trajectory, extend it outward by twice the vehicle length until an obstacle is encountered, and record the outer remaining distance Lout. If Lout < Lin, it means the outer distance is insufficient, and directly perform an overall incremental update. If Lout ≥ Lin, perform local dynamic correction; 33) Saving the scene time-sharing operation results: According to the above analysis and decision, execute the corresponding path correction measures, and save the time-sharing operation results after optimization processing as the basis for the next moment's decision; Step 4, Initialization of the A* algorithm architecture: Initialize the nodes of the A* algorithm and assign corresponding attributes, set the corresponding cost function g(n) and heuristic function h(n) for each node, set the iteration depth, and define a two-way optimization heuristic function; Step 5, Bidirectional iterative loop search for the path: Search for the path bidirectionally, gradually iterate and deepen the loop process, and merge the bidirectional paths to achieve rapid planning; Step 6, Scene dynamic update and path output: When the calculation state in Step 3 is local dynamic correction, adopt the local dynamic correction strategy. When the calculation state in Step 3 is overall incremental update, adopt the overall incremental update strategy. Respond to the dynamic changes of the scene through local dynamic correction or overall incremental update, and perform the final path output.

2. The method for dynamic scene planning of an intelligent vehicle based on time-sharing computing according to claim 1, characterized in that: The method for scene information processing and key element extraction in Step 1 includes: 11) Raw data acquisition and preprocessing: Real-time environmental data is obtained from vehicle-mounted sensors as raw data, denoising is performed on the raw data, and coordinate system conversion is performed to unify the data into the same reference coordinate system; 12) Determine the resolution of the grid map based on the actual application scenario: The resolution of the grid map should be as large as possible, but should be less than half the width of the vehicle, less than half the minimum turning radius of the vehicle, and less than the maximum distance the vehicle can travel in a perception and control cycle; 13) Extract key scene elements: Extract the position and size information of traffic participants and obstacles related to the movement of the main vehicle in the scene for updating the grid map. According to the sensor data, each grid is assigned an attribute value to represent the characteristics of the grid.

3. The method for dynamic scene planning of an intelligent vehicle based on time-sharing computing according to claim 1 is characterized in that: Step 2: The short-term motion prediction judgment and output method includes: 21) Construct a basic kinematic prediction model: A nonlinear kinematic model is used to describe vehicle motion. The basic structure of the model is as follows: ① The vehicle consists of two wheels, and the direction is controlled by the front wheel steering angle δ; ② The position of the vehicle's center of mass is represented by the coordinates (x, y), and the heading angle is represented by θ; the dynamic equation is as follows: Longitudinal dynamics include: Among them, v represents the vehicle speed, that is, the speed of the center of mass along the vehicle's forward direction, and a represents the vehicle acceleration; Lateral dynamics include: Among them, ω represents the vehicle yaw rate, β represents the steering slip angle, L is the wheelbase, and r is the turning radius; Using the relationship in the Frenet coordinate system, we can deduce: Combine the longitudinal dynamics and lateral dynamics equations to form a complete kinematic prediction model to calculate the vehicle state (x, y, θ, v, ω) at the future moment; 22) Kinematic short-term prediction output: Given the current vehicle state parameters (x(0), y(0), θ(0), v(0), ω(0)), the future driving instructions are obtained based on the historical state calculations. The vehicle state at the future time t is obtained by performing integral calculations based on the kinematic prediction model equations, including position, speed, and heading angle. The calculation formula is as follows: x(t)=x(0)+∫[v]dt y(t)=y(0)+∫[v×sin(θ+β)]dt θ(t)=θ(0)+∫[ω]dt v(t)=v(0)+∫[a]dt Thus, the vehicle trajectory data within the future preset time period is output, including the position (x, y) and posture (θ) information at each time node; 23) Scene grid mapping: Fill the output vehicle trajectory data into the rasterized map of step 1.

4. The method for dynamic scene planning of an intelligent vehicle based on time-sharing computing according to claim 1 is characterized in that: Step 4: A* algorithm architecture initialization method includes: 41) Initialize the starting node and the target node, and assign corresponding attributes, including coordinates and states; set the corresponding cost function and heuristic function for each node; 42) Construct data structure: create two open lists, corresponding to the forward search from the starting point to the end point and the reverse search from the end point to the starting point; at the same time, create two closed lists to record the information of the nodes that have been visited and processed to ensure that repeated calculations and invalid traversals are avoided during the search process; 43) Set the iteration depth and bidirectional optimization heuristic function: Set the initial iteration depth, and increase this limit by a certain value each time the iteration deepens; bidirectional optimization requires symmetry to ensure that the search process can be effectively guided in both directions.

5. The method for dynamic scene planning of an intelligent vehicle based on time-sharing computing according to claim 4 is characterized in that: Manhattan distance is used as the heuristic function to ensure the consistency of bidirectional search; For two points (x1, y1) and (x2, y2) on a two-dimensional plane, the Manhattan distance is defined as: h manhattan =|x1-x2|+|y1-y2|。 6. The method for dynamic scene planning of an intelligent vehicle based on time-sharing computing according to claim 1, characterized in that: Step 5: The bidirectional iterative loop search method of the path includes: 51) Bidirectional search execution: forward A search is performed from the starting point, and reverse A search is performed from the target node at the same time; in each round of iteration, the node with the highest priority is selected for expansion according to F(n)=g(n)+h(n); 52) Gradually iterate and deepen the loop process: When a round of search fails to find a solution or reaches the current iteration depth limit, increase the iteration depth limit and continue the next round of search; in each iteration, the open list and closed list must be maintained, and the cost information of the node must be updated; 53) Path intersection detection and merging: When the forward search front and the reverse search front meet, it means that a possible shortest path has been found; trace back to their respective start / target nodes in the forward and reverse directions to form two half paths, which are then spliced ​​together to obtain the complete optimal path.

7. The method for dynamic scene planning of an intelligent vehicle based on time-sharing computing according to claim 1 is characterized in that: Step 6: The scene dynamic update and path output method includes: 61) When the calculation status of step 3 is local dynamic correction, the local dynamic correction strategy is adopted, including: a. calculating the cost function change of the affected section and adjusting the G value and F value of the corresponding node according to the new environmental information; b. recalculating the shortest path of the affected part based on the existing path and performing local optimization; c. updating the open list and closed list to reflect the latest path status; 62) When the calculation state in step 3 is an overall incremental update, the overall incremental update strategy is adopted, including: a. resetting the starting point and target point according to the new environment information; b. resetting the search process and re-executing the bidirectional A* search according to the new iteration depth limit; c. in the new search process, if the global update is not caused by the scene cycle time sharing, using the known valid information to accelerate the search convergence; 63) Output the final path.

8. An intelligent automobile dynamic scene planning system based on time-sharing computing, characterized in that: Execute the method for dynamic scene planning of intelligent vehicles based on time-sharing operation described in claim 1; the system includes an information processing module, a short-term prediction module, a time-sharing operation module and a path planning module; the information processing module processes the scene information input by the sensor, performs road grid conversion, constructs a basic grid, and extracts key elements in the scene, specifically executing the operation of step 1 of claim 1; the short-term prediction module constructs a prediction model based on basic kinematics, predicts the phased parameters of the moving subject, fills the scene grid, and specifically executes the operation of step 2 of claim 1; the time-sharing operation module uses the prediction data and the scene cycle to identify the time-sharing characteristics, divides the scene into stages, and specifically executes the operation of step 3 of claim 1; the path planning module first initializes the nodes and iteration depth of the A* algorithm, defines a bidirectional optimization heuristic function, and then bidirectionally searches for paths, gradually iterates and deepens the cycle process, merges bidirectional paths to achieve rapid planning, and finally responds to dynamic changes in the scene through local dynamic correction and overall incremental update, and outputs the final path results, specifically executing the operations of steps 4 to 6 of claim 1; ultimately, the intelligent vehicle completes an efficient and safe driving process.

Images