Scenario-adaptive autonomous driving vehicle decision-making planning method, device and medium
By combining reinforcement learning and model prediction and control, dynamic traffic information is used to make decisions and trajectory planning of autonomous vehicles, decision-making problems in complex environments are solved and the safety and comfort of autonomous vehicles are improved.
Patent Information
- Application Number
- CN202210770893.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-30
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2042-06-30
AI Technical Summary
It is difficult for autonomous vehicles to make reasonable decisions in complex and dynamic traffic environments to balance safety, comfort and driving efficiency, and it is difficult for the existing technology to achieve adaptive adjustment of scenarios.
Adopting a reinforcement learning architecture, the pre-trained decision-making end output layer network, planning end output layer network and horizontal and vertical progressive trajectory planning network are used to make behavioral decisions and trajectory planning, combining model predictive control (MPC) and cost Function design optimizes steering wheel angle and throttle opening to realize the scene adaptability of autonomous driving cars.
It improves the decision-making and planning capabilities of autonomous vehicles in different scenarios, improves safety and comfort, and enhances adaptability to complex traffic environments.
Smart Images

Figure CN115195784B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of autonomous driving, and in particular to a scene-adaptive autonomous driving vehicle decision-making planning method, device and medium. Background Art
[0002] With the continuous advancement of science and technology, artificial intelligence has gradually entered people's lives, and its application in the field of autonomous driving has become more extensive. Driven by the development of related technologies, autonomous driving technology with cars as the actual carrier has also developed rapidly. Autonomous driving technology refers to the ability of a car to actively control the car through the coordination and cooperation of its own functional modules without the driver's operation, and complete the behavior of reaching the destination safely, efficiently and accurately from the departure point. The emergence of autonomous driving cars has greatly helped to reduce the driver's workload and make rational use of driving time. In addition, mature autonomous driving technology can effectively reduce the incidence of accidents caused by irregular driving of drivers, such as fatigue driving, drunk driving, etc. Therefore, it is necessary to improve the safety and comfort of autonomous driving technology.
[0003] However, as far as the current autonomous driving technology is concerned, there are still unresolved technical problems before it can achieve completely autonomous driving without driver operation. Autonomous driving vehicles have different requirements for driving behavior in different scenarios. Generally speaking, autonomous driving technology should first ensure the personal and property safety of drivers and passengers; in scenarios where their safety is not threatened, it should pursue the riding comfort of drivers and passengers, and at the same time pursue higher evaluation indicators related to the performance of the vehicle, including stability and driving efficiency. It is crucial to enable autonomous driving vehicles to quickly make reasonable decision-making behaviors in a random and complex dynamic traffic environment, make correct balance adjustments under multiple indicator requirements, and achieve adaptive adjustment of the scene. The current autonomous driving technology can ensure that the car can complete the driving task well in a specific traffic scenario, but the driving scenario in reality is uncertain, and the traffic road condition information actually fed back to the autonomous driving vehicle is updated in real time and is more complex. In this case, making reasonable behavioral decisions and trajectory planning for the changing traffic information to ensure the safety and comfort experience of drivers and passengers is a technical bottleneck that autonomous driving technology currently needs to break through. Summary of the invention
[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a scene-adaptive autonomous driving car decision planning method, equipment and medium.
[0005] The purpose of the present invention can be achieved by the following technical solutions:
[0006] A scenario-adaptive autonomous driving vehicle decision planning method is used to make decision planning for an autonomous driving vehicle, comprising the following steps:
[0007] Obtain dynamic traffic information for the autonomous vehicle to be planned;
[0008] Input dynamic traffic information into the pre-trained decision-end output layer network to obtain lane-changing decisions;
[0009] Input dynamic traffic information into the pre-trained planning output layer network to obtain the MPC parameter adjustment results for vehicle trajectory planning under the current working conditions;
[0010] The lane-changing decision and MPC parameter adjustment results are input into the pre-trained lateral and longitudinal progressive trajectory planning network, and the steering wheel angle and throttle opening are output to the actuator to realize the autonomous driving function.
[0011] Preferably, during the training process of the decision-end output layer network, the vehicle state information and dynamic traffic information at the current moment are used as the state, lane changing or lane keeping is used as the action, and the vehicle's safety, comfort and driving efficiency are used as rewards to train the reinforcement learning neural network, and the state of the next timestamp is returned to update the vehicle and dynamic traffic information. Through multiple rounds of training, the reward function value is accumulated, and the decision-end output layer network can obtain the lane changing decision as lane changing or lane keeping based on the dynamic traffic information according to the training results.
[0012] Preferably, during the training process of the output layer network at the planning end, the vehicle state information and dynamic traffic information at the current moment are used as the state, MPC parameter values in different proportions are used as actions, and the safety and comfort of the vehicle are used as rewards to train the reinforcement learning neural network, and the state of the next timestamp is returned to update the vehicle and dynamic traffic information. Through multiple rounds of training, the reward function value is accumulated, and the output layer network at the planning end obtains the MPC parameter adjustment result of the trajectory planning stage based on the dynamic traffic information according to the reverse training result.
[0013] Preferably, in the trajectory planning stage, model predictive control is adopted for the lateral and longitudinal motions of the autonomous driving vehicle, and the cost function is designed as follows:
[0014] J=Q*J desire +R u *J u +R Δu *J Δu +ρ*ε 2
[0015] Among them, the cost function of trajectory planning is:
[0016]
[0017] The cost function for controlling input and action changes is designed as:
[0018]
[0019] System constraint design includes:
[0020] Safe area constraints (X,Y) min ≤(X,Y) t+jt ≤(X,Y) max
[0021] Actuator Constraints nin ≤u t+jt ≤u max
[0022] Smoothness constraint Δu min ≤Δu t+jt ≤Δu max
[0023] Relaxation factor constraint 0≤ε≤ε max
[0024] Among them, J represents the optimization objective function value, J desire represents the cost function of trajectory planning, J u represents the cost function of the control input, J Δu represents the cost function of action change, ε represents the relaxation factor, Q, R u , R Δu , ρ represents the weight matrix, N p represents the prediction time interval, N c Indicates the control time interval, Represents the predicted value of the output variable for each timestamp in the prediction time domain at the current moment, Represents the expected value of the output variable for each time stamp in the prediction domain at the current moment, (X,Y) min Indicates the minimum boundary value of the horizontal and vertical displacement corresponding to the safety area requirements, (X, Y) t+jt Indicates the horizontal and vertical displacement values corresponding to each timestamp in the control time domain at the current moment, (X, Y) max Indicates the minimum boundary value of the lateral and longitudinal displacement of the vehicle's position corresponding to the safety area requirements. Indicates the value of the control variable acceleration at each time stamp in the control time domain at the current moment, u min 、u max represents the upper and lower limits of the control variable acceleration constraint, Indicates the change value of the control variable acceleration at each time stamp in the control time domain at the current moment, Δu min , Δu max Represents the change value of the control variable, that is, the upper and lower limits of the acceleration change constraint, ε maxIndicates the maximum value of the relaxation factor.
[0025] Preferably, the lateral and longitudinal progressive trajectory planning network includes a longitudinal trajectory planning layer, a lateral trajectory planning and motion control layer, and a feedforward longitudinal motion control layer.
[0026] Preferably, the input of the longitudinal trajectory planning layer is dynamic traffic information, and the output is the longitudinal acceleration information of the next timestamp of the autonomous driving vehicle and the expected speed information and expected lateral displacement information of the current timestamp. The optimization objective constrains the longitudinal displacement, speed, acceleration, rate of change of acceleration and following distance of the autonomous driving vehicle.
[0027] Preferably, the input of the lateral trajectory planning and motion control layer is the longitudinal acceleration information of the autonomous driving vehicle at the next timestamp and the expected speed information and expected lateral displacement information at the current timestamp output by the longitudinal trajectory planning layer, and the output is the steering wheel angle information of the autonomous driving vehicle at the next timestamp. The optimization target takes into account the requirements of the linear model, lateral stability, and ground adhesion, and adds constraints to the tire slip angle, lateral acceleration, and acceleration of the autonomous driving vehicle respectively.
[0028] Preferably, the input of the feedforward longitudinal motion control layer is the longitudinal acceleration information of the autonomous driving vehicle at the next timestamp output by the longitudinal trajectory planning layer, and the output is the desired throttle opening of the autonomous driving vehicle at the next timestamp.
[0029] An electronic device comprises: a processor; and a memory on which executable codes are stored. When the executable codes are executed by the processor, the processor executes the above method.
[0030] A non-transitory machine-readable storage medium stores executable code thereon. When the executable code is executed by a processor of an electronic device, the processor is caused to execute the above method.
[0031] Compared with the prior art, the present invention introduces a reinforcement learning architecture, makes full use of the dynamic traffic information provided by the perception system, and realizes the adaptive adjustment of the MPC parameters of the decision-making planning stage of the autonomous driving vehicle in different scenarios through the construction of reinforcement learning problems, so as to improve the scene adaptation ability of the autonomous driving vehicle, thereby effectively improving the safety and comfort of the autonomous driving vehicle. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 It is a flow chart of the algorithm architecture of the present invention;
[0033] Figure 2 A schematic diagram of the reinforcement learning neural network structure of the autonomous driving car in the present invention;
[0034] Figure 3 A schematic diagram of the reinforcement learning neural network structure of the autonomous driving car in the present invention;
[0035] Figure 4 It is a schematic diagram of the structure of the lateral and longitudinal progressive trajectory planning with longitudinal safety priority in the present invention;
[0036] Figure 5 Schematic diagram of lane changing decision of the autonomous driving vehicle in the present invention. DETAILED DESCRIPTION
[0037] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Note that the following embodiments are merely illustrative in nature, and the present invention is not intended to limit its applicable objects or uses, and the present invention is not limited to the following embodiments.
[0038] Example 1
[0039] A scenario-adaptive autonomous driving vehicle decision-making planning method, which includes two parts: training and application:
[0040] In the training part, the parameters of the introduced model predictive controller are adaptively adjusted through the reinforcement learning method under the actor-critic architecture. Specifically, the strategy deep neural network structure is designed in the form of output layer separating the output decision information and planning information, such as Figure 2 , 3 As shown in the figure, the model predictive control adopts a longitudinal and lateral automatic driving trajectory control structure that prioritizes longitudinal safety, as shown in the figure. Figure 4 As shown; the specific implementation of the above content is:
[0041] Training part 1, the output layer network of the decision-making end of the reinforcement learning neural network is used in the behavioral decision-making stage to determine whether the vehicle needs to execute a lane change decision under the current working conditions. Its operating mechanism is as follows: the current vehicle state information and dynamic traffic information are used as the state, "lane change" or "lane keeping" is used as the action, and the vehicle's safety, comfort, and driving efficiency are used as rewards to train the reinforcement learning neural network and return the state of the next timestamp, that is, update the vehicle and dynamic traffic information. Through multiple rounds of training, the reward function value is accumulated, and according to the training results, it can be determined under what working conditions the lane change decision needs to be executed. Among them, the vehicle state information refers to the lateral and longitudinal displacement and speed of the vehicle.
[0042] In the training part 2, the output layer network of the planning end of the reinforcement learning neural network is used in the trajectory planning stage to design the most appropriate MPC parameter settings for the vehicle in the current scenario. Its operating mechanism is as follows: the current vehicle state information and dynamic traffic information are used as the state, the MPC parameter values of different proportions are used as the action, and the safety and comfort of the vehicle are used as the reward to train the reinforcement learning neural network and return the state of the next timestamp, that is, to update the vehicle and dynamic traffic information. Through long-term training, the reward function value is accumulated, and according to the training results, the reasonable MPC parameters in the trajectory planning stage can be obtained for different scenarios, that is, the multi-scenario adaptive construction of the MPC parameters of the autonomous driving vehicle is realized. Among them, the vehicle state information refers to the lateral and longitudinal displacement and speed of the vehicle.
[0043] In training part 3, the solution process of the present invention adopts a transverse and longitudinal progressive trajectory planning network with longitudinal safety priority, including a longitudinal trajectory planning layer, a transverse trajectory planning and motion control layer, and is supplemented by a feedforward longitudinal motion control layer to increase its safety.
[0044] Training part 4, the longitudinal trajectory planning layer of model predictive control. The function of this layer is to keep the autonomous vehicle in the dynamic safety area and maintain the desired distance from the surrounding obstacle vehicles. Its input is the dynamic traffic information of the upper perception system, and its output is the longitudinal acceleration information of the next timestamp of the autonomous vehicle and the expected speed information and expected lateral displacement information of the current timestamp. The optimization goal constrains the longitudinal displacement, speed, acceleration, rate of change of acceleration and following distance of the autonomous vehicle.
[0045] Training part 5, lateral trajectory planning and motion control layer of model predictive control. The function of this layer is to realize the lateral expected position of the upper-layer longitudinal trajectory planning and complete the "lane keeping" or "lane change" instructions in the behavior decision stage. Its input is the longitudinal acceleration information of the next timestamp of the autonomous driving vehicle output by the longitudinal trajectory planning layer and the expected speed information and expected lateral displacement information of the current timestamp. The output is the steering wheel angle information of the autonomous driving vehicle at the next timestamp. The optimization goal takes into account the requirements of linear model, lateral stability, and ground adhesion, and adds constraints to the tire slip angle, lateral acceleration, and acceleration of the autonomous driving vehicle.
[0046] Training part 6, model predictive control feedforward longitudinal motion control layer. The function of this layer is to realize the control of the autonomous driving motion trajectory under the upper layer lateral and longitudinal progressive trajectory planning framework. Its input is the longitudinal acceleration information of the autonomous driving car at the next time stamp output by the longitudinal trajectory planning layer, and its output is the throttle opening expected at the next time stamp of the autonomous driving car.
[0047] The application part combines the trained reinforcement learning neural network and MPC algorithm with examples of different driving scenarios to obtain controller output and execute it to realize the function of autonomous driving car. The specific implementation is as follows:
[0048] Step 1: The environmental perception system of the autonomous driving vehicle uses sensors to obtain dynamic traffic information, that is, the dynamic traffic information includes the shape information (length, width, height), position information, speed information, and heading angle information of the obstacle vehicle.
[0049] Step 2: Input the dynamic traffic information obtained by the perception system into the reinforcement learning neural network. After calculation, the lane change decision is output, that is, whether to execute the lane change and issue a lane keeping or lane change command.
[0050] Step 3: Input the dynamic traffic information into the pre-trained planning end output layer network to obtain the MPC parameter adjustment results for vehicle trajectory planning under the current working conditions;
[0051] Step 4: Input the lane change decision and MPC parameter adjustment results into the pre-trained lateral and longitudinal progressive trajectory planning network, output the steering wheel angle and throttle opening to the actuator, and realize the function of the autonomous driving vehicle. In the trajectory planning stage, the MPC (model predictive control) method is adopted for the lateral and longitudinal movement of the autonomous driving vehicle, and its cost function is designed as follows:
[0052] J=Q*J desire +R u *J u +R Δu *J Δu +ρ*ε 2
[0053] Among them, the cost function of trajectory planning is:
[0054]
[0055] The cost function for controlling input and action changes is designed as:
[0056]
[0057] System constraint design includes:
[0058] Safe area constraints (X,Y) min ≤(X,Y) t+jt ≤(X,Y) max
[0059] Actuator Constraints min ≤u t+jt ≤u max
[0060] Smoothness constraint Δumin ≤Δu t+jt ≤Δu max
[0061] Relaxation factor constraint 0≤ε≤ε max
[0062] Among them, J represents the optimization objective function value, J desire represents the cost function of trajectory planning, J u represents the cost function of the control input, J Δu represents the cost function of action change, ε represents the relaxation factor, Q, R u , R Δu , ρ represents the weight matrix, N p represents the prediction time interval, N c Indicates the control time interval, Represents the predicted value of the output variable for each timestamp in the prediction time domain at the current moment, Represents the expected value of the output variable for each time stamp in the prediction domain at the current moment, (X,Y) min Indicates the minimum boundary value of the horizontal and vertical displacement corresponding to the safety area requirements, (X, Y) t+jt Indicates the horizontal and vertical displacement values corresponding to each timestamp in the control time domain at the current moment, (X, Y) max Indicates the minimum boundary value of the lateral and longitudinal displacement of the vehicle's position corresponding to the safety area requirements. Indicates the value of the control variable acceleration for each time stamp in the control time domain at the current moment, u min 、u max represents the upper and lower limits of the control variable acceleration constraint, Indicates the change in the acceleration of the control variable at each time stamp in the control time domain at the current moment, Δu min , Δu max Represents the change value of the control variable, that is, the upper and lower limits of the acceleration change constraint, ε max Indicates the maximum value of the relaxation factor.
[0063] The MPC parameters in the trajectory planning mentioned in step 3, namely Q, R u ,R Δu The equation J related to parameter Q is desire It represents the distance between the lateral position of the autonomous vehicle and the planned trajectory when the vehicle is driving. The large weight of Q indicates that the path tracking ability of the autonomous vehicle is emphasized under this condition, which is usually used as a safety evaluation indicator. u The relevant formula J u Represents the input control quantity of the autonomous vehicle, generally refers to the steering wheel angle, longitudinal acceleration, etc. uA large weight indicates that the state adjustment of the autonomous driving vehicle is of great concern under this working condition, which is usually used as an evaluation index of power and comfort. Δu The relevant formula J Δu Represents the rate of change of the input control quantity of the autonomous driving vehicle, R Δu A larger weight indicates that under this condition, there is more concern about whether the autonomous vehicle has sudden steering behavior and large acceleration changes, which is usually used as an evaluation indicator of comfort.
[0064] The following examples illustrate the adjustment of parameters under different working conditions. Figure 5 As shown, vehicle E is the ego vehicle and the others are traffic vehicles. When the decision behavior of the ego vehicle is to change lanes to the left, the parameter value needs to be adjusted for different working conditions. If the working conditions are: the speed of traffic vehicle No. 1 is greater than or equal to the ego vehicle and the longitudinal distance between it and the ego vehicle does not affect the lane-changing behavior of the ego vehicle, the speed of traffic vehicle No. 2 is less than or equal to the ego vehicle and will not accelerate, change lanes or perform other actions that affect the lane-changing behavior during the lane-changing process, the speed of traffic vehicle No. 3 is greater than or equal to the ego vehicle and the distance between it and the ego vehicle does not affect the lane-changing behavior of the ego vehicle, and other traffic vehicles do not have sudden speed changes or lane changes that affect the lane-changing behavior, that is, when the ego vehicle has sufficient lane-changing conditions, it has higher requirements for comfort and power, and the weight coefficient R should be appropriately increased. u , R Δu value; if the operating conditions are: vehicle No. 1 slows down or changes lanes to the right, vehicle No. 2 accelerates or has a small longitudinal distance from the vehicle, which increases the difficulty of changing lanes for the vehicle, vehicle No. 3 slows down, or other vehicles change speed or change lanes, which increases the difficulty of changing lanes, that is, when the lane changing behavior of the vehicle is potentially dangerous, the safety requirements will increase, and the value of the weight coefficient Q should be increased.
[0065] The trajectory planning control fusion method based on reinforcement learning and MPC mentioned in step 4, that is, the reinforcement learning training will be based on the MPC model to form a model-based reinforcement learning algorithm, which has the same goal as MPC and is committed to improving the safety and comfort of autonomous driving vehicles. And the reinforcement learning training is meaningful only when the MPC operation has results, that is, under the premise of meeting safety. On the one hand, this method reduces the difficulty of reinforcement learning training and improves the operating efficiency; on the other hand, it ensures the safety and executability of the entire architecture.
[0066] Example 2
[0067] An electronic device comprises: a processor; and a memory having executable code stored thereon, wherein when the executable code is executed by the processor, the processor executes a scene-adaptive autonomous driving vehicle decision-making planning method as described above.
[0068] A non-temporary machine-readable storage medium having executable code stored thereon, which, when executed by a processor of an electronic device, causes the processor to execute a scene-adaptive autonomous driving vehicle decision-making planning method as described above.
[0069] The implementation process of the functions and effects of each module in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, which will not be repeated here.
[0070] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this specification. A person of ordinary skill in the art can understand and implement it without paying creative labor.
[0071] The program product provided in the embodiments of the present application may adopt any combination of one or more readable media, wherein the readable medium may be a readable signal medium or a readable storage medium, and the readable storage medium may be but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any combination of the above. Specifically, more specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, RAM, ROM, Erasable Programmable Read Only Memory (EPROM), optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.
[0072] The program product provided in the embodiment of the present application may adopt a CD-ROM and include program code, and may also be run on a computing device. However, the program product provided in the embodiment of the present application is not limited thereto. In the embodiment of the present application, the readable storage medium may be any tangible medium containing or storing a program, and the program may be used by or in combination with an instruction execution system, apparatus, or device.
[0073] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.
[0074] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.
[0075] Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present application without departing from the spirit and scope of the embodiments of the present application. Thus, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.
[0076] The above embodiments are merely examples and do not limit the scope of the present invention. These embodiments can also be implemented in various other ways, and various omissions, substitutions, and changes can be made without departing from the technical concept of the present invention.
Claims
1. A scenario-adaptive autonomous driving vehicle decision-making planning method, used for making decision-making planning for an autonomous driving vehicle, characterized in that: The following steps are involved: Obtain dynamic traffic information for the autonomous vehicle to be planned; Input dynamic traffic information into the pre-trained decision-end output layer network to obtain lane-changing decisions; Input dynamic traffic information into the pre-trained planning output layer network to obtain the MPC parameter adjustment results for vehicle trajectory planning under the current working conditions; The lane change decision and MPC parameter adjustment results are input into the pre-trained lateral and longitudinal progressive trajectory planning network, and the steering wheel angle and throttle opening are output to the actuator to realize the function of autonomous driving car. During the training process of the decision-making end output layer network, the current vehicle state information and dynamic traffic information are used as the state, lane changing or lane keeping is used as the action, and the vehicle safety, comfort, and driving efficiency are used as rewards to train the reinforcement learning neural network, and the state of the next timestamp is returned to update the vehicle and dynamic traffic information. Through multiple rounds of training, the reward function value is accumulated, and the decision-making end output layer network obtains the lane changing decision based on the dynamic traffic information according to the training results, which is lane changing or lane keeping; During the training process of the output layer network at the planning end, the vehicle state information and dynamic traffic information at the current moment are used as the state, the MPC parameter values of different proportions are used as the action, and the safety and comfort of the vehicle are used as the reward to train the reinforcement learning neural network, and the state of the next timestamp is returned to update the vehicle and dynamic traffic information. Through multiple rounds of training, the reward function value is accumulated, and the output layer network at the planning end obtains the MPC parameter adjustment result of the trajectory planning stage based on the dynamic traffic information according to the training results; In the trajectory planning stage, model predictive control is adopted for the lateral and longitudinal motion of the autonomous driving vehicle, and the cost function is designed as follows: J=Q*J desire +R u *J u +R Δu *J Δu +ρ*ε 2 Among them, the cost function of trajectory planning is: The cost function for controlling input and action changes is designed as: System constraint design includes: Safe area constraints (X,Y min ≤(X,Y t+jt ≤(X,Y max Actuator Constraints min ≤u t+jt ≤u max Smoothness constraint Δu min ≤Δu t+jt ≤Δu max Relaxation factor constraint 0ε≤ε max Among them, J represents the optimization objective function value, J desire represents the cost function of trajectory planning, J u represents the cost function of the control input, J Δu represents the cost function of action change, ε represents the relaxation factor, Q, R u , R Δu , ρ represents the weight matrix, N p represents the prediction time interval, N c Indicates the control time interval, Represents the predicted value of the output variable for each timestamp in the prediction time domain at the current moment, Represents the expected value of the output variable for each time stamp in the prediction domain at the current moment, (X,Y) min Indicates the minimum boundary value of the horizontal and vertical displacement corresponding to the safety area requirements, (X, Y) t+kt Indicates the horizontal and vertical displacement values corresponding to each timestamp in the control time domain at the current moment, (X, Y) max Indicates the minimum boundary value of the lateral and longitudinal displacement of the vehicle's position corresponding to the safety area requirements. Indicates the value of the control variable acceleration at each time stamp in the control time domain at the current moment, u min 、u max represents the upper and lower limits of the control variable acceleration constraint, Indicates the change value of the control variable acceleration at each time stamp in the control time domain at the current moment, Δu min , Δu max Represents the change value of the control variable, that is, the upper and lower limits of the acceleration change constraint, ε max Indicates the maximum value of the relaxation factor.
2. The scenario-adaptive autonomous driving vehicle decision-making planning method according to claim 1, characterized in that: The transverse and longitudinal progressive trajectory planning network includes a longitudinal trajectory planning layer, a transverse trajectory planning and motion control layer, and a feedforward longitudinal motion control layer.
3. The scenario-adaptive autonomous driving vehicle decision-making planning method according to claim 2, characterized in that: The input of the longitudinal trajectory planning layer is dynamic traffic information, and the output is the longitudinal acceleration information of the next timestamp of the autonomous driving vehicle and the expected speed information and expected lateral displacement information of the current timestamp. The optimization objective constrains the longitudinal displacement, speed, acceleration, rate of change of acceleration and following distance of the autonomous driving vehicle.
4. The scenario-adaptive autonomous driving vehicle decision-making planning method according to claim 2, characterized in that: The input of the lateral trajectory planning and motion control layer is the longitudinal acceleration information of the next timestamp of the autonomous driving vehicle and the expected speed information and expected lateral displacement information of the current timestamp output by the longitudinal trajectory planning layer, and the output is the steering wheel angle information of the autonomous driving vehicle at the next timestamp. The optimization target takes into account the requirements of the linear model, lateral stability, and ground adhesion, and adds constraints to the tire slip angle, lateral acceleration, and acceleration of the autonomous driving vehicle respectively.
5. The scenario-adaptive autonomous driving vehicle decision-making planning method according to claim 2, characterized in that: The input of the feedforward longitudinal motion control layer is the longitudinal acceleration information of the autonomous driving vehicle at the next timestamp output by the longitudinal trajectory planning layer, and the output is the expected throttle opening of the autonomous driving vehicle at the next timestamp.
6. An electronic device, characterized in that: include: processor; and a memory having executable codes stored thereon, which, when executed by the processor, causes the processor to execute the method according to any one of claims 1 to 5.
7. A non-transitory machine-readable storage medium, characterized in that: An executable code is stored thereon, and when the executable code is executed by a processor of an electronic device, the processor is caused to execute the method as claimed in any one of claims 1 to 5.
Citation Information
Patent Citations
Intelligent vehicle transverse and longitudinal integrated control system and control method
CN110281912A
Intelligent driving automobile trajectory planning and tracking control method based on double-layer MPC
CN114312848A
