A human-machine collaborative driving simulation method, device, equipment and storage medium

Abstract

The present invention discloses a human-machine collaborative driving simulation method, device, equipment and storage medium, which relates to the field of driving simulation technology, and includes the following steps: building a simulation architecture for human-machine collaborative driving, including a simulated vehicle, a simulated environment, an improved two-point preview driver steering model and an automatic driving model, obtaining corresponding far-point preview targets and near-point preview targets through a preview style model, obtaining far-point errors and near-point errors according to the far-point preview targets and near-point preview targets; obtaining different degrees of lateral deviation of trajectories according to a driving proficiency model, and correcting the near-point error through different degrees of lateral deviation; inputting the far-point error and the corrected near-point error into the two-point preview driver steering model to obtain a first turning angle. The present invention increases the variable factors of the driver in the actual driving situation, so that the driver model can more accurately simulate the driver's situation in different states, and improves the accuracy of the final steering result.

Description

Technical Field

[0001] The present invention relates to the technical field of driving simulation, and in particular to a method, device, equipment and storage medium for human-machine collaborative driving simulation. Background Art

[0002] Human-machine collaborative driving refers to the process in which the driver and the autonomous driving system jointly intervene in the driving process from the perception layer, decision layer, and control layer through master-slave control, switching control, and shared control, thereby collaboratively completing the driving task. In driving tasks, drivers are good at reasoning and making decisions in complex environments, but their operating behaviors are easily affected by factors such as the environment and state, showing randomness and instability; while the driving system is good at refined perception and precise control, but its decision-making intelligence, learning ability, adaptability, and ability to understand complex environments are weaker than humans. Human-machine collaborative driving enables the driver and the driving system to cooperate with each other, give full play to their respective advantages, and make up for their respective shortcomings, thereby improving overall safety and decision-making capabilities.

[0003] As mentioned above, the core of human-machine collaborative driving lies in the reasonable allocation of driving rights. This mechanism requires human drivers and autonomous driving systems to participate in the driving process together. If this allocation strategy is to be studied, it is not feasible to directly use real drivers to drive vehicles equipped with autonomous driving systems on public roads for testing, because this not only poses a high safety risk, but also brings about large economic and time costs. Therefore, the best approach is to build a driver steering model that can well simulate the steering behavior of real drivers to replace the real driver, and integrate the model with the autonomous driving model into a high-fidelity simulation environment for research. This not only ensures the safety of the research, but also reduces costs, and can effectively evaluate the effects of different driving rights allocation strategies.

[0004] For the above-mentioned integrated environment, the main technical difficulties include the establishment of the driver steering model and the integration of the high-fidelity simulation environment. The driver steering model needs to accurately simulate the behavioral characteristics of real drivers, including but not limited to steering habits, reaction time, and decision-making patterns. However, due to the significant differences between individual drivers, how to design a universal and accurate model to cover the behavior of most people is a challenge. The high-fidelity simulation environment needs to be as close as possible to the complexity and uncertainty of the real world, including factors such as traffic flow, road conditions, and weather changes. How to ensure the accuracy and real-time nature of the simulation so that it can support real-time interaction and rapid iterative testing.

[0005] The existing common technical solutions use the Logitech G29 steering wheel kit, Simulink and CarSim vehicle dynamics software to build a human-machine collaborative driving test platform. Simulink provides a real-time simulation architecture, in which the intelligent vehicle autonomous steering control algorithm, driving rights allocation strategy, G29 input interface and visual dashboard are built; CarSim provides vehicle models, environmental models, sensors and visual scenes. It usually has the human-machine collaborative driving capabilities of manual driving, automatic driving and simultaneous human-machine in-loop.

[0006] However, although the test platform can complete the training and testing of human-machine collaborative driving, its simulation environment takes fewer factors into consideration and the working conditions are relatively simple, which results in the driver model being unable to accurately simulate the driver's situation under different conditions, causing the final steering result to deviate greatly from the actual situation. Summary of the invention

[0007] The present invention provides a human-machine collaborative driving simulation method, device, equipment and storage medium, which solves the problem that the existing method has fewer factors to consider in the simulation environment, the working conditions are relatively simple, and the driver model cannot accurately simulate the driver's situation under different conditions, resulting in a large deviation between the final steering result and the actual one.

[0008] In a first aspect, the present invention provides a human-machine collaborative driving simulation method, comprising the following steps:

[0009] A simulation framework for human-machine collaborative driving is constructed, wherein the simulation framework includes a simulated vehicle, a simulated environment, an improved two-point preview driver steering model and an automatic driving model; the improved two-point preview driver steering model adds a preview style model and a driving proficiency model for characterizing the driver in the actual driving process to the two-point preview driver steering model, wherein the preview style model is used to adjust the preview distance according to the speed information of the simulated vehicle and the curvature information of the road in the simulated environment to obtain a far-point preview target and a near-point preview target; the driving proficiency model is used to obtain the degree of lateral deviation of the trajectory of the simulated vehicle according to different proficiency levels;

[0010] When the simulated vehicle is driving in the simulation environment, the corresponding far-point preview target and near-point preview target are obtained through the preview style model, and the far-point error and near-point error are obtained according to the far-point preview target and the near-point preview target; different lateral deviation degrees of the trajectory are obtained according to the driving proficiency model, and the near-point error is corrected according to different lateral deviation degrees;

[0011] The far point error and the corrected near point error are input into a two-point preview driver steering model to obtain a first turning angle; the first turning angle is fused with a second turning angle output by the automatic driving model to obtain a fused turning angle, and the simulated vehicle steers according to the fused turning angle.

[0012] Preferably, the preview style model is used to adjust the preview distance according to the speed information of the simulated vehicle and the curvature information of the road in the simulated environment to obtain a far-point preview target and a near-point preview target, as shown below:

[0013]

[0014] Where η 1 , η 2 and η 3 To describe the parameters of different preview styles, |q| is the preview curvature threshold of the route, V is the vehicle speed, and d is the near-point preview target.

[0015] Preferably, the driving proficiency model is used to obtain the degree of lateral deviation of the trajectory of the simulated vehicle according to different proficiency levels, as shown below:

[0016]

[0017] in,

[0018]

[0019] Where Δe is the lateral deviation of the trajectory, |Δe| max is the maximum allowable lateral deviation, |Δe| min is the minimum perceptible lateral deviation value, α and μ are scaling factors, and W r is the road width, D is the simulated vehicle width

[0020] Preferably, obtaining the far point error and the near point error according to the far point preview target and the near point preview target comprises the following steps:

[0021] Obtain the angle between the current position of the simulated vehicle and the far-point preview target to obtain the far-point error;

[0022] Get the lateral distance between the current position of the simulated vehicle and the near-point preview target, and get the near-point error.

[0023] Preferably, the improved two-point preview driver steering model further includes a decision deviation characterization module, and the decision deviation characterization module is used to correct the first turning angle, as shown below:

[0024] ω * =(1+σ)ω;

[0025] In the formula, ω * is the actual first turning angle, (1+σ) is the driver’s actual decision interval for the lateral steering angle, ω is the first turning angle, and the actual first turning angle is fused with the second turning angle output by the autonomous driving model.

[0026] Preferably, a simulation environment for human-machine collaborative driving is built through the ROS system.

[0027] Preferably, the simulation architecture also includes a decision model, which is used to fuse the actual first turning angle with the second turning angle output by the autonomous driving model to obtain a fused turning angle.

[0028] In a second aspect, the present invention provides a human-machine collaborative driving simulation device, comprising:

[0029] A building module is used to build a simulation framework for human-machine collaborative driving, wherein the simulation framework includes a simulated vehicle, a simulated environment, an improved two-point preview driver steering model and an automatic driving model; the improved two-point preview driver steering model adds a preview style model and a driving proficiency model for characterizing the driver in the actual driving process to the two-point preview driver steering model, wherein the preview style model is used to adjust the preview distance according to the speed information of the simulated vehicle and the curvature information of the road in the simulated environment to obtain a far-point preview target and a near-point preview target; the driving proficiency model is used to obtain the lateral deviation degree of the trajectory of the simulated vehicle according to different proficiency levels;

[0030] The correction module is used to obtain the corresponding far-point preview target and near-point preview target through the preview style model when the simulated vehicle is driving in the simulation environment, and obtain the far-point error and near-point error according to the far-point preview target and the near-point preview target; obtain different lateral deviation degrees of the trajectory according to the driving proficiency model, and correct the near-point error according to different lateral deviation degrees;

[0031] The steering module is used to input the far point error and the corrected near point error into the two-point preview driver steering model to obtain a first turning angle; the first turning angle is merged with the second turning angle output by the automatic driving model to obtain a merged turning angle, and the simulated vehicle is steered according to the merged turning angle.

[0032] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned human-machine collaborative driving simulation method when executing the program.

[0033] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the human-machine collaborative driving simulation method described above is implemented.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention adds a preview style model and a driving proficiency model to the two-point preview driver steering model, wherein the preview style model is used to adjust the preview distance according to the speed information of the simulated vehicle and the curvature information of the simulated environment, and can dynamically adjust the preview distance according to the actual situation to more accurately reflect the driver's behavioral characteristics. The driving proficiency model is used to adjust the lateral deviation degree of the trajectory of the simulated vehicle according to different proficiency. The present invention increases the variable factors of the driver in the actual driving situation, so that the driver model can more accurately simulate the driver's situation under different conditions, and improve the accuracy of the final steering result. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 A hybrid decision-making architecture diagram of a human-machine collaborative driving simulation method of the present invention;

[0038] Figure 2 A schematic diagram of a driver steering model of the present invention;

[0039] Figure 3 This is a schematic diagram of a two-point preview of a driver of the present invention;

[0040] Figure 4 This is a structural diagram of the brain emotion learning circuit model of the present invention;

[0041] Figure 5 A diagram of the semicircular canal model of the vestibular perception system of the present invention;

[0042] Figure 6 It is a schematic diagram of the ROS node and Message transmission flow of the present invention;

[0043] Figure 7 It is a visualization schematic diagram of the near point and the far point in the Carla simulation environment of the present invention;

[0044] Figure 8 The present invention is a flowchart of a human-machine collaborative driving simulation method. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] The invention provides a human-machine collaborative driving simulation method, referring to Figure 8 , including the following steps:

[0047] Step 1: Build a simulation architecture for human-machine collaborative driving.

[0048] Reference Figure 1 The architecture includes: simulated vehicle, simulated environment, driver model, Autoware autonomous driving model and ROS2 (Robot Operating System 2).

[0049] Driver model: This model represents the behavior of human drivers, who usually focus on two key points on the road - near and far. The near point information is used to adjust the vehicle's driving direction in real time, while the far point information is used to predict the curve of the road ahead so that the driver can prepare in advance.

[0050] The Autoware autonomous driving model is an open source autonomous driving software platform that includes a series of modules, such as perception, positioning and planning. Its input is real-time data information of simulated vehicles to realize autonomous driving functions.

[0051] ROS2 (Robot Operating System 2): ROS2 is a robot operating system that provides underlying support and services for robot development. In this architecture, ROS2 acts as a bridge between the two-point preview driver model and Autoware, which exchange data through a publish / subscribe mechanism.

[0052] Human-machine collaborative driving hybrid decision model: This model can usually be a model to be verified. It attempts to combine the outputs of the two-point preview driver model and the Autoware model to achieve human-machine collaborative driving. This hybrid decision model can select appropriate driving strategies according to different scenarios. For example, it may rely more on human judgment in complex road conditions, and use more autonomous driving technology in simple road conditions.

[0053] Carla simulation environment and ROS Rviz: Carla is an open source driving simulation for testing and validating autonomous driving algorithms. ROS Rviz is a visualization tool for viewing and interactively editing ROS data streams.

[0054] Reference Figure 2 The driver model of the present invention is an improved two-point preview driver steering model. The two-point preview driver steering model is divided into two parts: a near-point decision model and a far-point decision model, which respectively correspond to the driver's near-point preview and far-point preview behaviors during driving.

[0055] Near-point decision model: The driver first extracts near-point information from the road, which may be a reference point near the current vehicle position. Then, the near-point information is converted into a form that can be processed by the brain's emotional learning loop model through the sensory input function. The brain's emotional learning loop model calculates an error signal based on the received information, and then applies fuzzy rules and learning rates to update the weights.

[0056] Far-point decision model: Similarly, the driver also extracts far-point information from the road, which is a reference point that may be passed in the future. The far-point information is also converted into a form that can be processed by the brain's emotional learning circuit model through the sensory input function. The brain's emotional learning circuit model calculates the error signal based on the received information, and then applies fuzzy rules and learning rates to update the weights.

[0057] The result of the decision-making turning angle will be given to the vehicle model through the arm model, and the driver's reflex response speed can be set in the arm model. The steering speed of the vehicle will be corrected by the vestibular model and fuzzy rules, and fed back to the forward proportional coefficient. Therefore, the preview ratio of the near point and the far point can be adjusted.

[0058] Through the interaction of these two models, the decision-making process of the driver in actual driving is simulated, including the preview behavior of near and far points. Through continuous learning and feedback, the system can gradually optimize its decision-making ability, thereby more accurately controlling the vehicle's motion state.

[0059] like Figure 3 As shown in the figure, the selected near point on the center line of the road can be used as a correction reference for the lateral deviation of the vehicle, ensuring that the vehicle continues to approach the center line of the road, thereby achieving accurate lateral tracking of the road trajectory. The position information of the far point as the tangent point helps the driver to estimate the curvature of the road ahead in advance. When performing steering maneuvers, the driver mainly relies on two reference points on the road, the near and far points, to obtain visual preview information. The preview near point information helps to continuously reduce the lateral deviation between the vehicle and the center line of the road, ensuring that the vehicle accurately tracks the road trajectory. The preview far point (i.e., tangent point) information enables the driver to estimate the curvature of the road ahead in advance, and accordingly estimate the size of the steering wheel angle, which is positively correlated with the curvature θ. By integrating the information of the near and far points, the driver can effectively control the steering of the vehicle.

[0060] like Figure 4As shown in the figure, the brain's emotional learning circuit is composed of multiple core components, including the thalamus, sensory cortex, orbitofrontal cortex, and amygdala. In this complex neural network, the thalamus plays a key role. It is responsible for receiving and processing a variety of sensory input signals SI from the external environment. Once the thalamus identifies the highest sensory signal intensity Ath, it immediately sends this highest value to the amygdala. To describe this process in mathematical language, we can get the following expression:

[0061] A th =max(SI i ).

[0062] It is then transmitted to the sensory cortex, where the sensory signal SI is further processed to generate reference cues for the emotional signal EC.

[0063] Next, the sensory signal SI enters the amygdala, where memory learning begins under the influence of the emotional signal EC, generating a learning signal A.

[0064] At the same time, the sensory signal SI is also sent to the orbitofrontal cortex and supervised by the emotional signal EC to correct the overall emotional learning circuit and finally produce the correction signal O:

[0065] E = AO.

[0066] The signal value output by the amygdala is tracking the emotional signal EC. When the output signal of the amygdala does not keep up with the emotional signal EC, the weight factor of the corresponding item will increase, and the direction of increase is the same as the sign direction of the sensory input signal SI. When the output signal value of the amygdala exceeds the emotional signal EC, the weight factor remains unchanged. The weight factor update process is described in mathematical language as:

[0067]

[0068] Each sensory input signal sent to the orbitofrontal cortex will find a corresponding node in the orbitofrontal cortex and associate with it. Each such node has a variable connection weight factor. When the sensory input signal is multiplied by the corresponding weight factor, the output value of the node is obtained.

[0069]

[0070] The output signal value of the orbitofrontal cortex reflects its inhibitory and corrective effects on the output value of the entire brain emotion learning circuit. This is an individual self-adjustment mechanism and a manifestation of reverse inhibition. When the output value of the brain emotion learning circuit exceeds the emotion signal EC, the orbitofrontal cortex will increase its weight factor, thereby enhancing its own output value, and then inhibiting the output of the entire brain emotion learning circuit. On the contrary, when the output value of the orbitofrontal cortex is lower than the emotion signal EC, it will reduce the weight factor and weaken its own output, thereby correcting the output of the entire brain emotion learning circuit. The weight factor update process is described in mathematical language as follows:

[0071]

[0072] Different drivers and the same driver may show unique preview habits in different situations, resulting in differences in the distances of the near and far points they focus on. Therefore, the improved two-point preview driver steering model adopts an adaptive preview experience index model (preview style model), which can dynamically adjust the preview distance according to the actual situation to more accurately reflect the driver's behavioral characteristics. The model is:

[0073]

[0074] Among them, η 1 ,η 2 ,η 3 Different values ​​represent different preview styles (η 1 ,η 2 ,η 3 It is a parameter that describes different preview styles. When observing the road ahead, the driver will dynamically adjust his preview distance according to the curvature and boundary conditions of the path. When facing a road with a large curvature, in order to enhance driving stability, the driver usually chooses a shorter preview distance; when the curvature is small, a longer preview distance is preferred. In addition, as the speed changes, the preview behavior will also be adjusted accordingly: when driving at high speed, the driver is more likely to take a long-distance preview; when driving at low speed, he prefers to observe at a closer distance. η 3 is the lower limit of the preview distance that the driver can take. |q| is the preview curvature threshold of the route, and V is the vehicle speed. The calculated distance d is the near-point preview target, and adding Δd is the far-point preview target. The degree of deviation is obtained by the preview target position and the current position. The degree of deviation is used as input to the driver model to generate the corresponding decision, and then the final decision is corrected by the decision deviation.

[0075] Different drivers may have different driving proficiency, which is reflected in the perception of preview error and the accuracy of actual control. Therefore, a personalized deviation model (driving proficiency model) and a decision deviation characterization module are introduced.

[0076] For the driving proficiency model, since different drivers have different distance control capabilities, their perception of preview errors will also be different. Specifically, if the lateral deviation of the vehicle's trajectory Δe can be expressed as:

[0077]

[0078] In the above formula, |Δe| max Indicates the maximum allowable lateral deviation value, |Δe| min Indicates the minimum perceived lateral deviation value, which can be adjusted by the actual road width and the driver's personal preference.

[0079]

[0080] In the above formula, α and μ represent the scaling factors, which can be adjusted according to the driver's preference. r is the road width and D is the vehicle width.

[0081] For the decision deviation characterization module, it is difficult to ensure that each control is a very accurate value in the actual decision-making process, but it is very likely to be another control value close to the accurate value. That is, the decision steering will produce a decision deviation from the actual steering. That is:

[0082] ω * =(1+σ)ω;

[0083] Where (1+σ) is the actual decision range of the driver for the lateral steering angle. Each steering decision will be different and fluctuate around 1. is the normal distribution of variance. That is: In the formula, is the driver's proficiency. The smaller the value, the more proficient the driver's actual control behavior is, and vice versa. * is the actual decision steering angle result.

[0084] When the vehicle is driving, the driver will constantly receive feedback from the outside world. In this process, the information perceived by the driver is somewhat different from the actual information. The vestibular perception system is an important part of the human body's perception of the external world, so this part is modeled.

[0085] Table 1 Model parameters

[0086] Pitch (around Y axis) Roll (around X axis) Yaw (around Z axis) <![CDATA[T L (s)]]> 5.3 6.1 10.2 <![CDATA[T s (s)]]> 0.1 0.1 0.1 <![CDATA[T a (s)]]> 30 30 30 <![CDATA[δ TH ((°) / s)]]> 3.6 3.0 2.6

[0087] like Figure 5 , ω is the yaw angular velocity input of the current vehicle carrier; ω qis the yaw rate perceived by the driver's nervous system. The parameters can be determined by the specific axis selection. The transfer function of the vestibular perception system can be summarized as:

[0088]

[0089] When the steering wheel is in the initial position and not rotating, if the driver holds the steering wheel in a relaxed state, then in this case, the stiffness and damping characteristics of the driver's arm are mainly determined by the mechanical properties of its muscles and skin. In this state, the interaction between the driver's arm and the steering wheel can be regarded as a spring damping system, in which the arm's response to the random disturbance torque applied by the steering wheel conforms to the dynamic response characteristics of the system. The transfer function can be expressed as follows:

[0090]

[0091] The parameters are selected as follows: Arm moment of inertia J dr =0.064kgm 2 , arm damping B dr =0.56Nms / rad, arm stiffness K dr =3.8Nms / rad. The steering system including the steering wheel, steering column, steering rack, gear, rack, and tire can also be regarded as a spring damping system. The moment of inertia of the system J st =0.172kgm 2 , damping B st =0.41Nms / rad, stiffness K st =3.8Nm / rad.

[0092] Step 2: When the simulated vehicle is driving in the simulation environment, the corresponding far-point preview target and near-point preview target are obtained through the preview style model, and the far-point error e is obtained based on the far-point preview target and the near-point preview target. y And the near-point error θ. According to the driving proficiency model, different trajectory lateral deviation degrees Δe are obtained, and the near-point error is corrected by different lateral deviation degrees.

[0093] Among them, the far point error is the angle error between the body direction of the current position of the simulated vehicle and the far point preview target, and the near point error is the error of the projection distance between the current position of the simulated vehicle and the near point preview target on the x-axis, that is, the lateral distance.

[0094] Step 3: Input the far point error and the corrected near point error into the two-point preview driver steering model to obtain the first turning angle. The first turning angle is fused with the second turning angle output by the autonomous driving model to obtain the fused turning angle, and the simulated vehicle steers according to the fused turning angle.

[0095] like Figure 6 As shown in the figure, the involved Carla vehicle sensor data will be reported to the ROS system, and then sent to the driver model and the autonomous driving model through the subscription and publishing mode. After the consumption calculation by the driver model node group and the autonomous driving model node group, the obtained results are transmitted to the hybrid decision node and the fusion execution node, and finally the control command is sent to the vehicle in Carla. Among them, the trajectory prediction node will load the waypoints of the specified position of the Carla built-in map, calculate the coordinates of the near point and the far point of the preview node, and calculate the control amount through the control node. The same is true for the autonomous driving model node group. The hybrid decision node can use different decision methods. For example, the simplest decision method is the average of the two angles; you can also build a reinforcement learning model to get a smoother and softer decision angle. After the execution node in the above figure receives the data, in order to simulate the real effect of turning the steering wheel in real life, the steering wheel angle will be gradually adjusted until the target angle is reached.

[0096] Figure 7 The visualization of near and far points in the Carla simulation environment is shown, which is very useful for understanding the changes in the driver's visual focus and its impact on vehicle control. In this way, researchers can intuitively observe how the driver adjusts the vehicle's driving trajectory based on near and far point information, and how this information affects the driver's emotions and decision-making. The advantages of doing so include:

[0097] Visualization can help people better understand the changes in the driver's visual focus and how these focus points affect the vehicle's driving trajectory. Through graphical representation, it is clear how the driver makes decisions based on near and far point information.

[0098] In the Carla simulation environment, researchers can set different road conditions and driving scenarios to study the performance of drivers in different situations. This helps to understand the behavior patterns of drivers in different environments and how they adapt to different road conditions.

[0099] Visualization makes it easy to collect and analyze data, such as how much attention a driver pays to near and far points, and how these attention points affect parameters such as the vehicle's trajectory and speed.

[0100] ROS Rviz is a powerful visualization tool that can display various sensor data, vehicle status, and other relevant information in the autonomous driving system. The advantages of doing so include:

[0101] ROS Rviz allows users to monitor the operating status of the autonomous driving system in real time, including parameters such as the vehicle's route, speed, acceleration, and data from sensors such as radar, camera, and lidar.

[0102] With ROS Rviz, engineers can quickly discover and solve problems that may arise in autonomous driving systems, such as abnormal sensor data and path planning errors.

[0103] ROS Rviz can help developers debug and optimize driving algorithms by displaying system status in real time, identifying potential problems and improving algorithms.

[0104] This architecture uses a more accurate preview model: the weights of the preview tangent point and near point can be adjusted according to the state of the vehicle, which more realistically simulates the driver's visual attention allocation and improves the accuracy of the model.

[0105] This architecture uses a smarter autonomous driving model: the Autoware system is introduced as an autonomous driving model to obtain better autonomous driving results.

[0106] This architecture uses the ROS system to support parallel computing: The ROS (Robot Operating System) system is used as a support to achieve the ability of parallel computing. As an open source robot operating system, ROS provides a rich set of modular components and tools to facilitate developers to perform distributed computing and multi-threaded programming. Through ROS, the new method can effectively distribute computing tasks between multi-core processors or multiple computers, speed up model training and simulation, and improve overall computing efficiency. In addition, the openness and community support of ROS also make the new method easier to integrate with other hardware devices and software systems, enhancing the flexibility and scalability of the system.

[0107] This architecture can use an efficient computing platform: using modern deep learning frameworks such as PyTorch or TensorFlow, combined with NVIDIA GPU for accelerated computing, greatly improves model training and simulation efficiency and reduces computing costs.

[0108] This architecture has a more friendly development environment: Compared with MATLAB, the deep learning framework used in the new method provides a more convenient debugging and development environment, lowers the development threshold, and avoids high software licensing fees.

[0109] This architecture has better scalability: the new method supports the operation of large-scale models, especially when performing high-fidelity simulations, without performance bottlenecks, which is conducive to coping with increasingly complex autonomous driving scenarios and task requirements.

[0110] Based on the same concept, the present invention also provides a human-machine collaborative driving simulation device, including a building module, a correction module and a steering module.

[0111] The building module is used to build a simulation architecture for human-machine collaborative driving. The simulation architecture includes a simulated vehicle, a simulated environment, an improved two-point preview driver steering model and an automatic driving model; the improved two-point preview driver steering model adds a preview style model and a driving proficiency model to the two-point preview driver steering model, which are used to characterize the driver's actual driving process. The preview style model is used to adjust the preview distance according to the speed information of the simulated vehicle and the curvature information of the road in the simulation environment to obtain the far-point preview target and the near-point preview target; the driving proficiency model is used to obtain the lateral deviation degree of the trajectory of the simulated vehicle according to different proficiency.

[0112] The correction module is used to simulate the driving process of the vehicle in the simulation environment. The corresponding far-point preview target and near-point preview target are obtained through the preview style model, and the far-point error and near-point error are obtained according to the far-point preview target and the near-point preview target; according to the driving proficiency model, different degrees of lateral deviation of the trajectory are obtained, and the near-point error is corrected according to different degrees of lateral deviation.

[0113] The steering module is used to input the far-point error and the corrected near-point error into the two-point preview driver steering model to obtain the first turning angle; the first turning angle is merged with the second turning angle output by the automatic driving model to obtain the fused turning angle, and the simulated vehicle steers according to the fused turning angle.

[0114] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned human-machine collaborative driving simulation method when executing the program.

[0115] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned human-machine collaborative driving simulation method is implemented.

[0116] Although the preferred embodiments of the present invention 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 that fall within the scope of the present invention.

[0117] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A human-machine collaborative driving simulation method, characterized in that: The following steps are involved: A simulation framework for human-machine collaborative driving is constructed, wherein the simulation framework includes a simulated vehicle, a simulated environment, an improved two-point preview driver steering model and an automatic driving model; the improved two-point preview driver steering model adds a preview style model and a driving proficiency model for characterizing the driver in the actual driving process to the two-point preview driver steering model, wherein the preview style model is used to adjust the preview distance according to the speed information of the simulated vehicle and the curvature information of the road in the simulated environment to obtain a far-point preview target and a near-point preview target; the driving proficiency model is used to obtain the degree of lateral deviation of the trajectory of the simulated vehicle according to different proficiency levels; When the simulated vehicle is driving in the simulation environment, the corresponding far-point preview target and near-point preview target are obtained through the preview style model, and the far-point error and near-point error are obtained according to the far-point preview target and the near-point preview target; different lateral deviation degrees of the trajectory are obtained according to the driving proficiency model, and the near-point error is corrected according to different lateral deviation degrees; Inputting the far point error and the corrected near point error into the two-point preview driver steering model to obtain the first turning angle; The first turning angle is fused with the second turning angle output by the automatic driving model to obtain a fused turning angle, and the simulated vehicle turns according to the fused turning angle.

2. A human-machine collaborative driving simulation method as claimed in claim 1, characterized in that: The preview style model is used to adjust the preview distance according to the speed information of the simulated vehicle and the curvature information of the road in the simulated environment to obtain the far-point preview target and the near-point preview target, as shown below: Where η1, η2 and η3 are parameters describing different preview styles, |q| is the distance preview curvature threshold, V is the vehicle speed, and d is the near-point preview target.

3. The human-machine collaborative driving simulation method according to claim 1, characterized in that: The driving proficiency model is used to obtain the lateral deviation degree of the trajectory of the simulated vehicle according to different proficiency levels, as shown below: in, Where Δe is the lateral deviation of the trajectory, |Δe| max is the maximum allowable lateral deviation, |Δe| min is the minimum perceptible lateral deviation value, α and μ are scaling factors, and W r is the road width, and D is the simulated vehicle width.

4. The human-machine collaborative driving simulation method according to claim 1, characterized in that: The method of obtaining a far point error and a near point error according to a far point preview target and a near point preview target comprises the following steps: Obtain the angle between the current position of the simulated vehicle and the far-point preview target to obtain the far-point error; Get the lateral distance between the current position of the simulated vehicle and the near-point preview target, and get the near-point error.

5. The human-machine collaborative driving simulation method according to claim 1, characterized in that: The improved two-point preview driver steering model further includes a decision deviation characterization module, and the decision deviation characterization module is used to correct the first turning angle, as shown below: oh * =(1+σ)ω; In the formula, ω * is the actual first turning angle, (1+σ) is the driver’s actual decision interval for the lateral steering angle, ω is the first turning angle, and the actual first turning angle is fused with the second turning angle output by the autonomous driving model.

6. The human-machine collaborative driving simulation method according to claim 1, characterized in that: A simulation environment for human-machine collaborative driving is built through the ROS system.

7. A human-machine collaborative driving simulation method as claimed in claim 5, characterized in that: The simulation architecture also includes a decision model, which is used to fuse the actual first turning angle with the second turning angle output by the autonomous driving model to obtain a fused turning angle.

8. A human-machine collaborative driving simulation device, characterized in that: include: A building module is used to build a simulation framework for human-machine collaborative driving, wherein the simulation framework includes a simulation vehicle, a simulation environment, an improved two-point preview driver steering model and an automatic driving model; the improved two-point preview driver steering model adds a preview style model and a driving proficiency model for characterizing the driver in the actual driving process to the two-point preview driver steering model, and the preview style model is used to adjust the preview distance according to the speed information of the simulation vehicle and the curvature information of the road in the simulation environment to obtain a far-point preview target and a near-point preview target; The driving proficiency model is used to obtain the degree of lateral deviation of the trajectory of the simulated vehicle according to different proficiency levels; The correction module is used to obtain the corresponding far-point preview target and near-point preview target through the preview style model when the simulated vehicle is driving in the simulation environment, and obtain the far-point error and near-point error according to the far-point preview target and the near-point preview target; obtain different lateral deviation degrees of the trajectory according to the driving proficiency model, and correct the near-point error according to different lateral deviation degrees; A steering module, used for inputting the far point error and the corrected near point error into a two-point preview driver steering model to obtain a first turning angle; The first turning angle is fused with the second turning angle output by the automatic driving model to obtain a fused turning angle, and the simulated vehicle turns according to the fused turning angle.

9. A computer device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the human-machine collaborative driving simulation method described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the human-machine collaborative driving simulation method described in any one of claims 1 to 7 is implemented.

Images