A method for detecting and controlling automatic driving of an intelligent networked vehicle and a control system thereof

Abstract

The present application relates to an automatic driving detection and control method for an intelligent networked vehicle and its control system, wherein the automatic driving detection and control method for an intelligent networked vehicle comprises: planning the target trajectory through a continuous reinforcement learning model based on steering decision information and lane information; judging whether the front wheel steering angle loss of the vehicle is greater than a threshold; if so, adjusting the target trajectory through the continuous reinforcement learning model. Compared with the related art, the automatic driving detection and control method for an intelligent networked vehicle provided by the present invention will consider the front wheel steering angle loss of the intelligent networked vehicle when planning the trajectory. When the front wheel steering angle loss is greater than the threshold, the planned target trajectory will be adjusted in time, thereby greatly reducing the probability of collision between the intelligent networked vehicle and the obstacle, solving the problem that the current intelligent networked vehicle is prone to collision with other vehicles during the steering process.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving, and in particular to an autonomous driving detection and control method for an intelligent connected vehicle and a control system thereof. Background Art

[0002] Intelligent connected vehicles, also known as Intelligent Connected Vehicle (ICV), refer to the organic combination of Internet of Vehicles and smart cars, and eventually a new generation of cars that can replace human operation. This type of car is equipped with advanced on-board sensors, controllers, actuators and other devices, and integrates modern communication and network technologies to realize intelligent information exchange and sharing between cars and people, cars, roads, and background. Intelligent connected vehicles are safe, comfortable, energy-saving, and efficient, and can provide drivers with a more convenient and safe travel experience. The technical basis of intelligent connected vehicles mainly includes Internet of Vehicles communication technology, Internet of Things technology, and artificial intelligence technology. Internet of Vehicles communication technology uses wireless communication technologies such as 3G / 4G / 5G and WIFI to realize communication between cars, cars and roads, and cars and clouds; Internet of Things technology realizes smart transportation by connecting with infrastructure, environmental sensors, road signs, etc.; artificial intelligence technology uses deep learning, computer vision and other technologies to realize functions such as automatic driving, vehicle diagnosis, and smart travel.

[0003] The automatic driving system of intelligent connected vehicles is mainly divided into two parts: trajectory planning and trajectory control. Among them, the trajectory planning part mainly plans the ideal path of the intelligent connected vehicle based on road information and obstacle information, and the trajectory control mainly controls the vehicle according to the planned ideal trajectory. The two work together to realize the driving control of the intelligent connected vehicle. However, trajectory control cannot ensure that the vehicle drives completely according to the planned ideal path. When the actual front wheel turning angle of the vehicle is greatly different from the ideal front wheel turning angle, it is easy to collide with other vehicles. Therefore, the current intelligent connected vehicles are prone to collision with other vehicles during the turning process.

[0004] There is currently no effective solution to the problem that current intelligent connected vehicles are prone to collision with other vehicles during turning. Summary of the invention

[0005] The present invention provides an intelligent network-connected vehicle automatic driving detection control method and a control system thereof to solve the problem that the current intelligent network-connected vehicle is prone to collision with other vehicles during steering.

[0006] In a first aspect, the present invention provides a method for detecting and controlling automatic driving of an intelligent networked vehicle, the method comprising:

[0007] Based on the steering decision information and the lane information, planning the target trajectory through a continuous reinforcement learning model;

[0008] Determining whether the front wheel turning angle loss of the vehicle is greater than a threshold;

[0009] If so, the target trajectory is adjusted through the continuous reinforcement learning model.

[0010] In some embodiments, determining whether the front wheel turning angle loss of the vehicle is greater than a threshold comprises:

[0011] Based on the target trajectory, obtaining an estimated value of the center of mass sideslip angle, an estimated value of the yaw angular velocity, and an estimated value of the lateral velocity of the vehicle through an unscented Kalman filter;

[0012] Obtaining an actual value of the center of mass sideslip angle, an actual value of the yaw angular velocity, and an actual value of the lateral velocity of the vehicle;

[0013] The loss factor of the vehicle is calculated by a target calculation model, and the loss factor is used to characterize the front wheel turning angle loss. The target calculation model is:

[0014] flag=0.7*|Δβ|+0.1*|Δω|+0.2*|Δv y |

[0015] Where Δβ represents the difference between the estimated value of the center of mass sideslip angle and the actual value of the center of mass sideslip angle, Δω represents the difference between the estimated value of the yaw angular velocity and the actual value of the yaw angular velocity, and Δv y It represents the difference between the estimated lateral velocity and the actual lateral velocity;

[0016] It is determined whether the loss factor is greater than a threshold.

[0017] In some of the embodiments, the threshold of the loss factor is between 0.4 and 0.6.

[0018] In some embodiments, based on the target trajectory, obtaining the estimated value of the center of mass sideslip angle, the estimated value of the yaw angular velocity, and the estimated value of the lateral velocity of the vehicle by unscented Kalman filtering includes:

[0019] Based on the target trajectory, a planned value of the center of mass sideslip angle, a planned value of the yaw rate, and a planned value of the lateral speed of the vehicle are obtained through a vehicle dynamics model;

[0020] Obtaining a front wheel steering angle planning value output by a vehicle controller according to the target trajectory;

[0021] Based on the planned value of the center of mass sideslip angle, the planned value of the yaw angular velocity, the planned value of the lateral velocity and the planned value of the front wheel turning angle, an estimated value of the center of mass sideslip angle, an estimated value of the yaw angular velocity and an estimated value of the lateral velocity of the vehicle are obtained through unscented Kalman filtering.

[0022] In some embodiments, planning the target trajectory through a continuous reinforcement learning model based on the turning decision information and the lane information includes:

[0023] Based on the steering decision information, the lane information, and the pitch angle and roll angle of the vehicle, the target trajectory is planned through a continuous reinforcement learning model.

[0024] In some embodiments, the method further comprises:

[0025] Obtaining obstacle information and lane information of the vehicle;

[0026] Processing the obstacle information through fuzzy rules to obtain the turning intention information of the vehicle;

[0027] The turning intention information and the lane information are processed by a discrete reinforcement learning model to obtain the turning decision information.

[0028] In some embodiments, the obstacle information of the vehicle includes first relative distance information and first relative speed information between a first obstacle and the vehicle, second relative distance information and second relative speed information between a second obstacle and the vehicle, and third relative distance information and third relative speed information between a third obstacle and the vehicle.

[0029] The first obstacle is the nearest obstacle on the left lane of the vehicle, the second obstacle is the nearest obstacle on the lane where the vehicle is located, and the third obstacle is the nearest obstacle on the right lane of the vehicle;

[0030] The turning intention information includes left turn intention information and right turn intention information.

[0031] In some of the embodiments, the fuzzy rules include a left-turn fuzzy rule and a right-turn fuzzy rule;

[0032] The step of processing the obstacle information by using fuzzy rules to obtain the turning intention information of the vehicle includes:

[0033] Processing the first relative distance information, the first relative speed information, the second relative distance information and the second relative speed information by using the left-turn fuzzy rule to obtain the left-turn intention information;

[0034] The second relative distance information, the second relative speed information, the third relative distance information and the third relative speed information are processed by the right-turn fuzzy rule to obtain the right-turn intention information.

[0035] In a second aspect, the present invention provides an automatic driving detection and control device for an intelligent networked vehicle, the device comprising:

[0036] A trajectory planning module, for planning the target trajectory through a continuous reinforcement learning model based on the steering decision information and the lane information;

[0037] A loss judgment module, used to judge whether the front wheel turning angle loss of the vehicle is greater than a threshold;

[0038] The trajectory planning module is also used to adjust the target trajectory through the continuous reinforcement learning model when the front wheel steering angle loss is greater than a threshold.

[0039] In a third aspect, the present invention provides an intelligent connected vehicle automatic driving detection and control system, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the intelligent connected vehicle automatic driving detection and control method described in the first aspect.

[0040] Compared with related technologies, the automatic driving detection and control method of the intelligent networked vehicle and its control system provided by the present invention take into account the front wheel steering angle loss of the intelligent networked vehicle when planning the trajectory. When the front wheel steering angle loss is greater than the threshold, the planned target trajectory will be adjusted in time, thereby greatly reducing the probability of collision between the intelligent networked vehicle and obstacles, solving the problem that the current intelligent networked vehicle is prone to collision with other vehicles during the steering process.

[0041] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flow chart of a method for detecting and controlling automatic driving of an intelligent networked vehicle provided in an embodiment of the present invention;

[0043] Figure 2 A comparison diagram of vehicle driving trajectories using different trajectory planning methods in one embodiment of the present invention;

[0044] Figure 3 is an obstacle avoidance effect diagram of the dynamic programming method in the verification embodiment of the present invention;

[0045] Figure 4 is an obstacle avoidance effect diagram of the reinforcement learning planning method in the verification embodiment of the present invention without considering the front wheel angle loss;

[0046] Figure 5 It is a partial flow chart of the automatic driving detection and control method of an intelligent connected vehicle provided in some embodiments of the present invention. DETAILED DESCRIPTION

[0047] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0048] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the", "these" and the like in this application do not represent quantitative restrictions, and they can be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether directly or indirectly. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, "A and / or B" can mean: A exists alone, A and B exist at the same time, and B exists alone. Usually, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0049] In an embodiment of the present invention, a method for detecting and controlling automatic driving of an intelligent network-connected vehicle is provided. Figure 1 is a flow chart of a method for detecting and controlling automatic driving of an intelligent networked vehicle provided in an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0050] Step S110, planning the target trajectory through a continuous reinforcement learning model based on the turning decision information and the lane information.

[0051] Step S120, determining whether the front wheel turning angle loss of the vehicle is greater than a threshold; if so, executing step S130.

[0052] Step S130, adjusting the target trajectory through a continuous reinforcement learning model.

[0053] The above steps are mainly performed by the local trajectory planner in the intelligent connected vehicle system. The turning decision information is provided by the behavior planner in the intelligent connected vehicle system, which is used to determine whether the intelligent connected vehicle turns left into the left lane or turns right into the right lane, and is further divided into left turn decisions and right turn decisions. Lane information includes lane-related information, such as the centerline position of each lane, the lane where the vehicle is located, and other information. The target trajectory is the local trajectory of the vehicle during the turning process. The planning of the target trajectory specifically includes the planning of the vehicle's driving speed and acceleration, heading angle and other states, and the target trajectory contains the above vehicle state information. The continuous reinforcement learning model can use the TD3 (Twin Delayed Deep Deterministic policy gradient algorithm) algorithm.

[0054] Specifically, in trajectory planning, the center line of the lane that the intelligent connected vehicle needs to turn into can be determined as a reference line, a Frenet coordinate system can be established, and then the midpoint of the target trajectory can be determined based on the Frenet coordinate system. For example, when the intelligent connected vehicle needs to turn left, the center line of the left lane can be determined as a reference line, and then the end point range can be determined within the left lane, and the end point of the target trajectory can be determined within the range.

[0055] The key point of the above scheme is that after the target trajectory planning is completed, the local trajectory planner will also calculate the front wheel angle loss of the vehicle in real time. The front wheel angle loss refers to the actual front wheel angle of the vehicle not reaching the planned value. For example, in the target trajectory, the front wheel angle of the intelligent connected car should be 10°, but the actual front wheel angle of the intelligent connected car may be only 8°, then the front wheel angle loss is 20%. In theory, if the intelligent connected car drives completely according to the target trajectory, it will not collide with other vehicles, because the target trajectory is planned for obstacle avoidance. The greater the front wheel angle loss, the greater the difference between the actual trajectory of the intelligent connected car and the target trajectory, and the easier it is to collide with other vehicles. Therefore, a threshold value of the front wheel angle loss can be set. When the front wheel angle loss is greater than the threshold, it is considered that the intelligent connected car has a greater collision risk and is in an uncontrollable state, and the target trajectory needs to be adjusted. For example, the threshold value of the front wheel angle loss can be set to 50%. Because at the same driving speed, the smaller the front wheel angle, the longer the longitudinal distance required for a complete lane change, and it is easier to collide with the obstacle ahead. Therefore, the target trajectory adjustment mainly reduces the driving speed in the target trajectory, thereby reducing the longitudinal distance required for a complete lane change.

[0056] In summary, the automatic driving detection and control method for intelligent networked vehicles provided by the present invention takes into account the front wheel steering angle loss of the intelligent networked vehicle when planning the trajectory. When the front wheel steering angle loss is greater than the threshold, the planned target trajectory will be adjusted in time, thereby greatly reducing the probability of collision between the intelligent networked vehicle and obstacles, solving the problem that the current intelligent networked vehicle is prone to collision with other vehicles during the turning process.

[0057] In one experiment, in a multi-obstacle environment, a steering angle deviation failure was set in the vehicle, resulting in the actual front wheel steering angle of the vehicle only reaching 50% of the expected steering angle. A dynamic programming method, a reinforcement learning method without considering the fault, and a reinforcement learning method with considering the fault (the trajectory planning method adopted by the control method of the present invention) were compared.

[0058] Figure 2 FIG. 1 is a comparison diagram of vehicle driving trajectories of different trajectory planning methods in one embodiment of the present invention. Figure 2 It can be seen that when the dynamic programming method is used, the ego vehicle exceeds the lane boundary after driving 100 meters. When the reinforcement learning method without considering the corner deviation loss is used, the ego vehicle fails to successfully avoid the last obstacle and fails to drive this time. When the reinforcement learning method considering the corner deviation loss is used, after the ego vehicle detects the corner deviation loss, the ego vehicle will take a deceleration action, which will produce a large pitch angle at the beginning, and then the ego vehicle will be in a relatively low speed state, so that the ego vehicle will travel a shorter distance to complete the lane change when changing lanes.

[0059] In some embodiments, step S120, determining whether the front wheel steering angle loss of the vehicle is greater than a threshold, includes:

[0060] Step S121, based on the target trajectory, obtain the estimated value of the center of mass sideslip angle, the estimated value of the yaw angular velocity and the estimated value of the lateral velocity of the vehicle through unscented Kalman filtering.

[0061] Step S122, obtaining the actual value of the center of mass sideslip angle, the actual value of the yaw angular velocity and the actual value of the lateral velocity of the vehicle.

[0062] Step S123, calculating the loss factor of the vehicle through the target calculation model. The loss factor is used to characterize the front wheel turning angle loss. The target calculation model is:

[0063] flag=0.7*|Δβ|+0.1*|Δω|+0.2*|Δv y |

[0064] Where Δβ represents the difference between the estimated value of the center of mass sideslip angle and the actual value of the center of mass sideslip angle (center of mass sideslip angle error), Δω represents the difference between the estimated value of the yaw angular velocity and the actual value of the yaw angular velocity (yaw angular velocity error), and Δv yIndicates the difference between the estimated lateral velocity and the actual lateral velocity (lateral velocity error).

[0065] Step S124, determining whether the loss factor is greater than a threshold value. If so, executing step S130.

[0066] Specifically, in the actual driving process of intelligent connected vehicles, the front wheel steering loss may be difficult to measure directly. In this case, the front wheel steering loss can be equivalent to the loss factor by calculating the difference between the estimated value and the actual value of the center of mass sideslip angle, yaw rate and lateral velocity, and calculating the loss factor through the target calculation model. There is a positive correlation between the loss factor and the front wheel steering loss. Through experiments, it can be seen that the relationship between the two is as follows:

[0067] Table 1 Relationship between front wheel turning angle loss and maximum loss factor

[0068]

[0069] From the data in the table, we can see that when the front wheel steering angle loss is between 10% and 40%, the maximum value of the loss factor is basically proportional to the loss of the front wheel steering angle, and the vehicle can be considered controllable at this time. When the front wheel steering angle loss reaches 50%, the maximum value of the loss factor changes suddenly. Based on the comprehensive data, when it is determined that the front wheel steering angle loss is still controllable within 50%, the threshold of the loss factor can be set between 0.4 and 0.6, such as 0.4, 0.5 or 0.6. At this time, the corresponding threshold of the front wheel steering angle loss is between 40% and 50%. Therefore, when the loss factor is greater than the threshold, it is determined that it will have a non-negligible impact on vehicle safety, and the target trajectory needs to be adjusted.

[0070] In some embodiments, step S121, based on the target trajectory, obtaining the estimated value of the center of mass sideslip angle, the estimated value of the yaw angular velocity, and the estimated value of the lateral velocity of the vehicle through unscented Kalman filtering, includes:

[0071] Based on the target trajectory, the vehicle dynamics model is used to obtain the planned value of the center of mass sideslip angle, the planned value of the yaw rate and the planned value of the lateral speed; the front wheel steering angle planned value output by the vehicle controller according to the target trajectory is obtained; based on the planned value of the center of mass sideslip angle, the planned value of the yaw rate, the planned value of the lateral speed and the planned value of the front wheel steering angle, the unscented Kalman filter is used to obtain the estimated value of the center of mass sideslip angle, the estimated value of the yaw rate and the estimated value of the lateral speed of the vehicle.

[0072] In this embodiment, a specific method for determining the estimated value of the center of mass sideslip angle, the estimated value of the yaw rate, and the estimated value of the lateral speed of the vehicle is provided. Specifically, the planned value of the center of mass sideslip angle, the planned value of the yaw rate, the planned value of the lateral speed, and the planned value of the front wheel turning angle can be used as inputs of the unscented Kalman filter model, and the estimated value of the center of mass sideslip angle, the estimated value of the yaw rate, and the estimated value of the lateral speed can be obtained through calculation of the unscented Kalman filter model.

[0073] In the planning of the target trajectory, the vehicle posture of the intelligent connected vehicle, namely the pitch angle and roll angle, can also be considered, so that a target trajectory that maintains the stability of the vehicle body can be planned.

[0074] Therefore, in some embodiments, step S110, planning the target trajectory through a continuous reinforcement learning model based on the turning decision information and the lane information, includes:

[0075] Based on the steering decision information, lane information, and the pitch and roll angles of the vehicle, the target trajectory is planned through a continuous reinforcement learning model.

[0076] Specifically, the reward function of the continuous reinforcement learning model during training is:

[0077]

[0078]

[0079]

[0080] r d =-3*|d|

[0081] r v =-0.5*|Δv|

[0082]

[0083] Where θ represents the pitch angle, represents the roll angle, d represents the distance loss at the end point of the trajectory, Δv represents the speed loss, a=0 represents that the intelligent connected vehicle does not collide or run out of the lane, and a=1 represents that the intelligent connected vehicle collides or runs out of the lane.

[0084] The distance loss at the end of the trajectory refers to the lateral distance between the end of the trajectory and the center line of the turning lane, and the speed loss refers to the difference between the planned speed and the target speed.

[0085] Through the above reward function, the continuous reinforcement learning model can learn to plan the target trajectory while stabilizing the pitch angle and roll angle as much as possible.

[0086] As mentioned above, the automatic driving detection and control method for intelligent networked vehicles based on reinforcement learning provided by the present invention has been described through multiple embodiments. The automatic driving detection and control method for intelligent networked vehicles described above mainly has two characteristics. First, the front wheel steering angle deviation is considered during trajectory planning. When the front wheel steering angle deviation is greater than the threshold, the target trajectory will be readjusted. Second, the pitch angle and roll angle of the vehicle are considered during trajectory planning, so that the vehicle has a more stable pitch angle and roll angle when driving along the target trajectory. The effectiveness of the above effects is illustrated by some verifications as follows.

[0087] In order to further verify that the proposed reinforcement learning-based automatic driving detection and control method for intelligent connected vehicles (reinforcement learning planning method considering front wheel steering angle loss) can safely avoid obstacles when the vehicle has steering angle deviation loss and has a stable body posture during driving, hardware-in-the-loop experiments are carried out to establish an environment that contains both static and dynamic obstacle vehicles.

[0088] 1. Set the vehicle to drive normally to verify that under the same environment, the designed intelligent networked vehicle automatic driving detection and control method has a smaller pitch angle and roll angle than the dynamic programming method. The experimental results are shown in the following table:

[0089] Table 2 Comparison of pitch angle and roll angle during normal driving

[0090]

[0091] It can be seen from the table that the automatic driving detection and control method for intelligent connected vehicles provided by the present invention can reduce the pitch angle and roll angle of the vehicle body, while the roll angle and pitch angle are more stable.

[0092] 2. The turning angle deviation loss is set to 50%, that is, the actual turning angle is only half of the expected turning angle, in order to verify that the proposed method can safely avoid obstacles after the turning angle deviation loss occurs and stabilize the vehicle body posture during driving.

[0093] The obstacle avoidance effect of the dynamic programming method is as follows Figure 3 , the obstacle avoidance effect of the reinforcement learning planning method without considering the front wheel angle loss is as follows Figure 4 . It can be seen from the figure that after the corner deviation loss occurs, the vehicle will exceed the road boundary using the dynamic programming method, and the vehicle will collide with the obstacle vehicle using the reinforcement learning planning method that does not consider the front wheel corner deviation loss. Both methods failed to successfully drive to the end. The intelligent networked vehicle automatic driving detection and control method provided by the present invention will decelerate at the beginning, causing a pitch impact, but the fluctuation range of the roll is minimal. The comparison table of the pitch angle and roll angle of each method is as follows:

[0094] Table 3 Comparison of pitch angle and roll angle when front wheel steering angle is lost

[0095]

[0096] It can be seen from the table that the reinforcement learning planning method considering the corner loss is more stable in body roll control.

[0097] Reference Figure 5 In some embodiments, the automatic driving detection and control method of an intelligent networked vehicle further includes:

[0098] Step S140, obtaining obstacle information and lane information of the vehicle.

[0099] Step S150, the obstacle information is processed by fuzzy rules to obtain the turning intention information of the vehicle.

[0100] Specifically, the obstacle information of the vehicle includes the first relative distance information and the first relative speed information between the first obstacle and the vehicle, the second relative distance information and the second relative speed information between the second obstacle and the vehicle, and the third relative distance information and the third relative speed information between the third obstacle and the vehicle; wherein the first obstacle is the nearest obstacle on the left lane of the vehicle, the second obstacle is the nearest obstacle on the lane where the vehicle is located, and the third obstacle is the nearest obstacle on the right lane of the vehicle; the turning intention information includes the left turn intention information and the right turn intention information. The fuzzy rules include the left turn fuzzy rule and the right turn fuzzy rule.

[0101] This step specifically includes:

[0102] Step S151, processing the first relative distance information, the first relative speed information, the second relative distance information and the second relative speed information through the left turn fuzzy rule to obtain the left turn intention information; Step S152, processing the second relative distance information, the second relative speed information, the third relative distance information and the third relative speed information through the right turn fuzzy rule to obtain the right turn intention information.

[0103] Step S160, processing the turning intention information and lane information through a discrete reinforcement learning model to obtain turning decision information.

[0104] In this embodiment, the above three steps are mainly performed by the behavior planner in the intelligent networked vehicle system, and the above steps can be performed to obtain the steering decision information for achieving obstacle avoidance. When avoiding obstacles, there are many factors to consider, such as the state information of the vehicle and the surrounding vehicles, including speed, distance, lane, etc. This means that the more obstacle vehicles there are around the vehicle, the more state inputs of the discrete reinforcement learning model, which will make the training time of the discrete reinforcement learning model longer. In order to improve the training speed of the discrete reinforcement learning model, the obstacle information is first processed by fuzzy rules to obtain the steering intention information of the vehicle. The steering intention information includes the left turn intention value and the right turn intention value, and then the left turn intention value and the right turn intention value are used as the state input of the discrete reinforcement learning model, which can greatly reduce the number of state inputs of the discrete reinforcement learning model and improve the model training speed.

[0105] For example, taking the left lane and the lane where the vehicle is located as an example, if the obstacle in front is closer and the obstacle on the left is farther away, the vehicle has a high willingness to turn left; if the speed of the obstacle in front is slower than that of the vehicle and the speed of the obstacle on the left is faster than that of the vehicle, the vehicle also has a high willingness to turn left. The combination of the two can give a higher willingness value for turning left. The four state quantities can then be simplified into one state quantity.

[0106] Among them, the discrete reinforcement learning model can adopt the DQN (Deep Q-Network) algorithm model. In a verification, the discrete reinforcement learning DQN model without fuzzy processing can reach the maximum number of training actions in each subsequent training after nearly 4,000 times, which means that the vehicle can complete the whole journey without collision and the intelligent connected car has completed the training. After adding fuzzy processing, the discrete reinforcement learning DQN model only needs about 3,000 times to complete the training, which proves that adding fuzzy processing can improve the training speed of discrete reinforcement learning models.

[0107] Specifically, in the left-turn fuzzy rule, the first relative distance information and the first relative speed information are both non-positively correlated with the left-turn intention information, and the second relative distance information and the second relative speed information are both non-negatively correlated with the left-turn intention information; in the right-turn fuzzy rule, the second relative distance information and the second relative speed information are both non-positively correlated with the right-turn intention information, and the third relative distance information and the third relative speed information are both non-negatively correlated with the right-turn intention information.

[0108] In the fuzzy rules, the input and output are divided into multiple fuzzy subsets from small to large. Different input fuzzy subsets have corresponding relationships with different output fuzzy subsets, and the corresponding relationships can be set according to the actual scenario. In the present invention, the fuzzy subsets of multiple inputs (the relative distance and relative speed between the vehicle and obstacles on different lanes, a total of four inputs) are associated with the fuzzy subset of one output (the willingness to turn left or right). Exemplarily, the relative distance between the vehicle and the obstacle can be set to range from 0 meters to 60 meters, and the relative speed (obstacle speed-vehicle speed) can be set to range from -10m / s to 10m / s, that is, the domain of the relative distance is [0, 60], and the domain of the relative speed is [-10, 10]. If a certain value is higher or lower than the limit value, it is taken as the limit value. For example, there is no obstacle on a lane, and the relative distance of the lane is taken as 60m, and each state variable is divided in its respective domain by a triangular membership function. The relative distance is divided into five fuzzy subsets, representing very close, close, medium, far, and very far distances; the relative speed is divided into five fuzzy subsets, representing very slow, slow, medium, fast, and very fast speeds; the turning intention is divided into five fuzzy subsets, representing very low, low, medium, high, and very high turning possibilities. Finally, after defuzzification using the centroid method, the left turn intention value and the right turn intention value of the vehicle are output.

[0109] In an embodiment of the present invention, a smart connected vehicle trajectory planning system is also provided, including:

[0110] The behavior planner is used to execute steps S140 to S160 in the above embodiment.

[0111] The local trajectory planner is used to execute steps S110 to S130 in the above embodiment.

[0112] In an embodiment of the present invention, a smart connected vehicle automatic driving detection and control device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. The terms "module", "unit", "subunit", etc. used below may implement a combination of software and / or hardware of predetermined functions. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0113] The intelligent network-connected vehicle automatic driving detection and control device provided in an embodiment of the present invention includes:

[0114] The trajectory planning module is used to plan the target trajectory through a continuous reinforcement learning model based on the steering decision information and lane information.

[0115] The loss judgment module is used to judge whether the front wheel turning angle loss of the vehicle is greater than a threshold.

[0116] The trajectory planning module is also used to adjust the target trajectory through a continuous reinforcement learning model when the front wheel steering loss is greater than a threshold.

[0117] The above-mentioned intelligent networked vehicle automatic driving detection and control device will consider the front wheel steering angle loss of the intelligent networked vehicle when planning the trajectory. When the front wheel steering angle loss is greater than the threshold, the planned target trajectory will be adjusted in time, thereby greatly reducing the probability of collision between the intelligent networked vehicle and obstacles, solving the problem that the current intelligent networked vehicle is prone to collision with other vehicles during the turning process.

[0118] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0119] In an embodiment of the present invention, an intelligent connected vehicle automatic driving detection and control system is also provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the intelligent connected vehicle automatic driving detection and control method provided by the present invention.

[0120] It should be understood that the specific embodiments described herein are only used to explain the application, rather than to limit it. Based on the embodiments provided in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of this application.

[0121] Obviously, the drawings are only some examples or embodiments of the present application. For ordinary technicians in the field, the present application can also be applied to other similar situations based on these drawings without creative work. In addition, it is understandable that although the work done in this development process may be complicated and lengthy, for ordinary technicians in the field, certain changes in design, manufacturing or production based on the technical content disclosed in this application are only conventional technical means and should not be regarded as insufficient content disclosed in this application.

Claims

1. A method for detecting and controlling automatic driving of an intelligent networked vehicle, characterized in that: The method comprises: Based on the steering decision information and lane information, the target trajectory is planned through a continuous reinforcement learning model; Determine whether the front wheel steering angle loss of the vehicle is greater than a threshold; If yes, adjusting the target trajectory through the continuous reinforcement learning model; Wherein, judging whether the front wheel turning angle loss of the vehicle is greater than a threshold comprises: Based on the target trajectory, obtaining an estimated value of the center of mass sideslip angle, an estimated value of the yaw angular velocity, and an estimated value of the lateral velocity of the vehicle through an unscented Kalman filter; Obtaining an actual value of the center of mass sideslip angle, an actual value of the yaw angular velocity, and an actual value of the lateral velocity of the vehicle; The loss factor of the vehicle is calculated by a target calculation model, and the loss factor is used to characterize the front wheel turning angle loss. The target calculation model is: in, It represents the difference between the estimated value of the center of mass sideslip angle and the actual value of the center of mass sideslip angle. represents the difference between the estimated yaw rate and the actual yaw rate. It represents the difference between the estimated lateral velocity and the actual lateral velocity; It is determined whether the loss factor is greater than a threshold, where the threshold of the loss factor is between 0.4 and 0.

6.

2. The method for detecting and controlling automatic driving of an intelligent networked vehicle according to claim 1, characterized in that: Based on the target trajectory, an estimated value of the center of mass sideslip angle, an estimated value of the yaw angular velocity, and an estimated value of the lateral velocity of the vehicle are obtained by unscented Kalman filtering, including: Based on the target trajectory, a planned value of the center of mass sideslip angle, a planned value of the yaw rate, and a planned value of the lateral speed of the vehicle are obtained through a vehicle dynamics model; Obtaining a front wheel steering angle planning value output by a vehicle controller according to the target trajectory; Based on the planned value of the center of mass sideslip angle, the planned value of the yaw angular velocity, the planned value of the lateral velocity and the planned value of the front wheel turning angle, an estimated value of the center of mass sideslip angle, an estimated value of the yaw angular velocity and an estimated value of the lateral velocity of the vehicle are obtained through unscented Kalman filtering.

3. The method for detecting and controlling automatic driving of an intelligent networked vehicle according to claim 1, characterized in that: Based on the steering decision information and the lane information, the target trajectory is planned through a continuous reinforcement learning model, including: Based on the steering decision information, the lane information, and the pitch angle and roll angle of the vehicle, the target trajectory is planned through a continuous reinforcement learning model.

4. The method for detecting and controlling automatic driving of an intelligent networked vehicle according to claim 1, characterized in that: The method further comprises: Obtaining obstacle information and lane information of the vehicle; Processing the obstacle information through fuzzy rules to obtain the turning intention information of the vehicle; The turning intention information and the lane information are processed by a discrete reinforcement learning model to obtain the turning decision information.

5. The method for detecting and controlling automatic driving of an intelligent networked vehicle according to claim 4, characterized in that: The obstacle information of the vehicle includes first relative distance information and first relative speed information between a first obstacle and the vehicle, second relative distance information and second relative speed information between a second obstacle and the vehicle, and third relative distance information and third relative speed information between a third obstacle and the vehicle. The first obstacle is the nearest obstacle on the left lane of the vehicle, the second obstacle is the nearest obstacle on the lane where the vehicle is located, and the third obstacle is the nearest obstacle on the right lane of the vehicle; The turning intention information includes left turn intention information and right turn intention information.

6. The method for detecting and controlling automatic driving of an intelligent networked vehicle according to claim 5, characterized in that: The fuzzy rules include a left-turn fuzzy rule and a right-turn fuzzy rule; The step of processing the obstacle information by using fuzzy rules to obtain the turning intention information of the vehicle includes: Processing the first relative distance information, the first relative speed information, the second relative distance information and the second relative speed information by using the left-turn fuzzy rule to obtain the left-turn intention information; The second relative distance information, the second relative speed information, the third relative distance information and the third relative speed information are processed by the right-turn fuzzy rule to obtain the right-turn intention information.

7. An automatic driving detection and control device for an intelligent network-connected vehicle, characterized in that: The device comprises: The trajectory planning module is used to plan the target trajectory through a continuous reinforcement learning model based on the steering decision information and lane information; A loss judgment module is used to judge whether the front wheel turning angle loss of the vehicle is greater than a threshold; The trajectory planning module is further used to adjust the target trajectory through the continuous reinforcement learning model when the front wheel angle loss is greater than a threshold; Determining whether the front wheel turning angle loss of the vehicle is greater than a threshold includes: Based on the target trajectory, obtaining an estimated value of the center of mass sideslip angle, an estimated value of the yaw angular velocity, and an estimated value of the lateral velocity of the vehicle through an unscented Kalman filter; Obtaining an actual value of the center of mass sideslip angle, an actual value of the yaw angular velocity, and an actual value of the lateral velocity of the vehicle; The loss factor of the vehicle is calculated by a target calculation model, and the loss factor is used to characterize the front wheel turning angle loss. The target calculation model is: in, It represents the difference between the estimated value of the center of mass sideslip angle and the actual value of the center of mass sideslip angle. represents the difference between the estimated yaw rate and the actual yaw rate. It represents the difference between the estimated lateral velocity and the actual lateral velocity; It is determined whether the loss factor is greater than a threshold, where the threshold of the loss factor is between 0.4 and 0.

6.

8. An intelligent network-connected vehicle automatic driving detection and control system, comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to execute the automatic driving detection and control method for an intelligent connected vehicle according to any one of claims 1 to 6.

Images