WO2021036083A1

WO2021036083A1 - Driver behavior model development method and device for automatic driving, and storage medium

Info

Publication number: WO2021036083A1
Application number: PCT/CN2019/123508
Authority: WO
Inventors: 杜光辉; 袁雁城; 张尧文
Original assignee: 格物汽车科技(苏州)有限公司
Priority date: 2019-08-26
Filing date: 2019-12-06
Publication date: 2021-03-04
Also published as: CN110497914B; CN110497914A

Abstract

Disclosed are a driver behavior model development method and device for automatic driving, and a storage medium. The method comprises the following steps: (1) establishing a basic model in a simulation environment according to a transitive relationship between an excitation input and a theoretical output; (2) using the same excitation input as that for establishing the basic model to intervene in a target vehicle in a field test scenario, so as to acquire real output data of the target vehicle under the excitation input in the field test scenario, wherein the real output data comprises target vehicle driver decision data and target vehicle operation change data; and (3) using the acquired real output data of the target vehicle in the field test scenario to correct the basic model. The present invention is different from a method for training a huge amount of data to acquire a driver behavior model on the basis of AI technological analysis in the prior art. A driver behavior model approximating real driver behavior to the greatest extent is acquired, and an end-to-end algorithm and an end-to-end solution are effectively extracted.

Description

Development method, equipment and storage medium of driver behavior model for autonomous driving

Technical field

The present invention relates to the technical field of automatic driving, in particular to a method, equipment and storage medium for developing a driver behavior model for automatic driving.

Background technique

Autonomous vehicles, also known as unmanned vehicles, are intelligent vehicles that realize unmanned driving through computer systems. They rely on artificial intelligence, visual computing, radar, monitoring devices, and global positioning systems to work together to enable the vehicle’s computer system to Automatically and safely maneuver motor vehicles without unmanned operation.

At present, some Internet companies and automobile manufacturers rely on artificial intelligence or deep learning methods to obtain prediction and decision-making model algorithms for autonomous vehicles. The realization theory is: through vehicle sensors (for example, video cameras, radar sensors, and laser rangefinders, etc.) ) Obtain road data in the driving scene, obtain driver behavior data from the driving process of the vehicle, thereby forming big data covering the scene and driver behavior, analyze the scene and driver behavior data through AI, and continuously iterate the algorithm accordingly. Will get an E2E solution from perception to execution. However, the actual situation is that both Waymo and domestic OEMs have collected a large amount of data, and none of them have extracted effective end-to-end solution algorithms and solutions. The main reasons are the following two points: First, every scene It is unique, and it is impossible to enumerate all the scenes produced at every moment on the earth in number; second, AI analysis cannot evaluate the behavior of surrounding vehicles based on the data collected by sensors (for example, whether it is safe, efficient, compliant, etc.) , Even scene classification is difficult to do.

Summary of the invention

The embodiments of the present invention provide a driver behavior model development method, device and storage medium for automatic driving, so as to solve the problem that the driver behavior model developed in the existing automatic driving technology cannot effectively extract the end-to-end solution algorithm and solution problem.

In order to solve the above technical problems, the present invention provides a driver behavior model development method for autonomous driving, which includes the following steps:

(1) Create a basic model in the simulation environment based on the transfer relationship between the excitation input and the theoretical output;

(2) Intervene the target vehicle in the field test scene using the same excitation input as the basic model to obtain the actual output data of the target vehicle under the excitation input in the field test scene. The actual output data includes the target vehicle driver's decision data and Operational change data of the target vehicle;

(3) Use the acquired real output data of the target vehicle in the field test scenario to correct the basic model.

In a preferred embodiment of the present invention, it further includes step (3), where the correction of the basic model includes:

Determine whether the actual output data of the target vehicle in the field test scenario is different from the theoretical output, if so, use the actual output data to update the theoretical output data, and use the updated theoretical output data to modify the basic model.

In a preferred embodiment of the present invention, it further includes that the excitation input includes operating change parameters of surrounding vehicles.

In a preferred embodiment of the present invention, the incentive input further includes road traffic information parameters and environmental information parameters.

In a preferred embodiment of the present invention, it further includes that the environmental information parameters include one or any combination of illuminance, weather, temperature, humidity, wind direction, and wind speed; the road traffic information parameters include traffic information parameters for main roads in urban areas, suburban areas One or any combination of road traffic information parameters, national road traffic information parameters, provincial road traffic information parameters, and high-speed traffic information parameters.

In a preferred embodiment of the present invention, the incentive input further includes the driver's age parameter, gender parameter, physiological parameter, psychological parameter, and driving age parameter.

In a preferred embodiment of the present invention, further including step (2), the target vehicle operation change data includes vehicle position change data, speed change data, acceleration change data, steering angle change data, yaw rate change data, lateral acceleration One or any combination of change data, accelerator opening change data, and brake pedal opening change data.

In a preferred embodiment of the present invention, it further includes the operating change parameters of the surrounding vehicles including vehicle position change data, speed change data, acceleration change data, steering angle change data, yaw rate change data, lateral acceleration change data, One or any combination of accelerator opening change data and brake pedal opening change data.

In order to solve the above technical problems, the present invention also provides a driver behavior model development device, which includes a memory, a processor, and a program stored in the memory and capable of running on the processor. When the processor is executed, the above-mentioned driver behavior model development method for automatic driving is realized.

In order to solve the above technical problems, the present invention also provides a computer-readable storage medium, the computer-readable storage medium stores a driver behavior model development program, which is implemented when the driver behavior model development program is executed by a processor. The above-mentioned driver behavior model development method for autonomous driving.

The driver behavior model development method, equipment, and storage medium for automatic driving disclosed in the embodiments of the present application are completely different from the prior art method based on AI technology analysis and training to obtain a driver behavior model from massive amounts of data. First, create a basic model in the simulation environment based on the transfer relationship between the excitation input and the theoretical output under the excitation input. Second, use the excitation input to intervene in the target vehicle in the field test scenario to obtain the true value of the target vehicle in the field test scenario. Output, and finally, use the real output in the field test scenario to modify the basic model to obtain a driver behavior model that is close to the real driver behavior to the greatest extent, and effectively extract end-to-end algorithms and solutions.

Description of the drawings

Fig. 1 is a flowchart of a method for developing a driver behavior model in a first embodiment of the present invention;

Figure 2 is the model diagram under the first field test scenario;

Figure 3 is a model diagram under the second field test scenario;

Figure 4 is a model diagram under the third field test scenario;

Fig. 5 is a structural block diagram of a driver behavior model development device in the second embodiment of the present invention.

detailed description

The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention, but the examples cited are not intended to limit the present invention.

This embodiment discloses a driver behavior model development method for automatic driving. As shown in FIG. 1, the development method includes the following steps:

(1) Create a basic model in the simulation environment based on the transfer relationship between the excitation input and the theoretical output.

The theoretical output here is the driving behavior predicted by the creator of the basic model to have 4 to 10 years of driving experience, and to have good driving behavior within the driving age, and the driving behavior of an experienced driver in the driving environment of the incentive input.

The excitation input here includes one or any combination of the following:

(A) Operating variation parameters of vehicles located around the target vehicle in the driving environment (including the front and rear directions of the same lane, the parallel position of the left and right adjacent lanes, and the front and rear directions of the left and right adjacent lanes). The operating change parameters of the surrounding vehicles here include vehicle position change data, speed change data, acceleration change data, steering angle change data, yaw rate change data, lateral acceleration change data, accelerator opening change data, brake pedal opening change One or any combination of data.

(B) Road traffic information parameters and environmental information parameters. The road traffic information parameters here include one or any combination of main road traffic information parameters in urban areas, suburban road traffic information parameters, national road traffic information parameters, provincial road traffic information parameters, and high-speed traffic information parameters; the environmental information parameters here include One or any combination of illuminance, weather, temperature, humidity, wind direction, wind speed.

(C) Driver's age parameters, gender parameters, physiological parameters, psychological parameters and driving age parameters.

(D) Vehicle type parameters.

The creator of the basic model creates the basic model in the simulation environment according to the transfer relationship between the excitation input and the theoretical output under the excitation input.

(2) Intervene the target vehicle in the field test scenario using the same excitation input as the basic model to obtain the actual output data of the target vehicle under the excitation input in the field test scenario. The actual output data includes the target vehicle driver's decision data and the target Vehicle operation change data. Here the target vehicle operation change data includes vehicle position change data, speed change data, acceleration change data, steering angle change data, yaw rate change data, lateral acceleration change data, accelerator opening change data, brake pedal opening change data One of them or any combination.

The key here is the setting of field test scenarios: design field test scenarios based on experience and typical scenes of Chinese roads collected throughout the year, including key elements such as road types, vehicle types, weather conditions, vehicle distribution conditions, and vehicle behaviors. Before the test starts, a test outline is prepared according to the needs of the driver's behavior model development. The outline content includes the definition of collected data, the change mode of vehicle operation in a single scene, and the change mode of vehicle operation in combined scenes. During the test, the person in charge of the test instructs the test vehicle (or "surrounding vehicle") to change its operation mode according to the requirements of the test program, as an incentive to collect the response mode of the driver of the tested vehicle (or "target vehicle"), including the driver The decision-making data and the operational change data of the tested vehicle.

(3) Use the acquired real output data of the target vehicle in the field test scenario to correct the basic model. The correction process here includes judging whether the actual output data of the target vehicle in the field test scenario is different from the theoretical output, if so, using the real output data to update the theoretical output data, and using the updated theoretical output data to modify the basic model.

Refer to Figure 2 for a non-threatening car following scene model:

In the field test SC1 scenario, the vehicle is controlled to accelerate or decelerate by sensing the speed of the preceding vehicle, the distance between the accelerometer, and other related information to ensure that the vehicle and the preceding vehicle maintain a safe and comfortable distance dx.

In the field test SC2 scenario, even if the vehicle in front does a certain degree of emergency braking without any warning, the vehicle can still maintain a certain safe distance ds with the vehicle in front at an acceptable deceleration.

Here, the own vehicle needs to obtain at least the following data: ①The speed of the preceding vehicle in the last time period; ②The acceleration of the preceding vehicle in the last time period; ③The distance between the own vehicle and the preceding vehicle in the last time period; ④The current vehicle in front Estimated parking distance at vehicle speed; ⑤Estimated parking distance at current speed of the vehicle. And output at least the following data: ①The distance between the vehicle and the preceding vehicle in the current time period; ②The acceleration of the vehicle in the current time period; ③The speed of the vehicle in the current time period.

The SC1 scene of the field test includes the correction of the difference between the acceleration and deceleration strategy, and the limitation of the experience value (or "threshold experience value") difference correction while maintaining the distance between the vehicle and the preceding vehicle. For example, in the field test SC1 scenario, the vehicle adopts the deceleration strategy, and the acceleration strategy is used when creating the basic model. In this scenario, the deceleration strategy is used to update the acceleration strategy, and one of the incentive input in this scenario and the updated acceleration strategy is used. The transfer relationship between updates the basic model.

The field test SC2 scenario includes the correction of the deceleration strategy difference and the deceleration experience value difference correction adopted by the vehicle after the emergency braking of the vehicle in front. For example, in the SC2 scene of the field test, the deceleration of the vehicle is a1, and when the basic model is created, the deceleration of the vehicle under the excitation input is a2 (a2 is not equal to a1), then the deceleration is updated with deceleration a2 in this scenario a1, and use the transfer relationship between the excitation input in this scenario and the updated deceleration a2 to update the basic model.

Refer to Figure 3 for a car following scene model under the threat of close CUT-IN:

In the field test SC3 scenario, the intention of changing lanes is predicted by sensing the trajectory of abnormal vehicles in adjacent lanes, and by actively shortening the distance between the vehicle and the preceding vehicle to avoid the full rights of hidden dangers caused by the arbitrary lane change of the adjacent vehicle , The vehicle and the preceding vehicle will eventually maintain a safe and compact distance dx.

In the field test SC4 scenario, even if the vehicle in front performs a certain degree of emergency braking without any warning, the vehicle can also change lanes to avoid it, and maintain a certain safe distance from the front and rear vehicles in the new lane. d _FR , d _RR .

Here, the vehicle needs to obtain at least the following data: ①The speed change and acceleration change range of the N time period before the lane change vehicle; ②The distance to the vehicle ahead and the change range of the N time period before the lane change vehicle and the change range; ③Last time Cycle distance between the vehicle in front and the vehicle in front; ④Estimation of the parking distance of the vehicle in front at the current speed; ⑤The usable area of adjacent lanes

And output at least the following data: ①The distance between the own vehicle and the preceding vehicle in the current time period; ②The acceleration of the own vehicle in the current time period; ③The speed of the own vehicle in the current time period; ④The transverse direction of the own vehicle in the lane in the current time period position.

The field test SC3 scenario includes the correction of the empirical value difference of the distance between the vehicle and the preceding vehicle at different speeds of the vehicle in front, and the revision of the difference of the change rule function.

The field test SC3 scenario includes the correction of the lateral control strategy difference, the correction of the experience value of the yaw rate, the correction of the change rate, etc. after the vehicle in front of the vehicle has braked at different levels.

Refer to Figure 4 for the model of the car's active lane change and overtaking scene

In the field test SC5 scenario, the vehicle is controlled longitudinally and laterally by sensing the speed, acceleration, and distance of the preceding vehicle, the vehicle being overtaken, and the neighboring vehicle to ensure that the vehicle and the neighboring vehicle stay in place. A safe and comfortable distance (-dx and dRR, etc.), even if the vehicle in front of the vehicle in front of the vehicle does a certain degree of emergency braking without any warning, the vehicle can safely change lanes to the expected adjacent lane.

Here, the own vehicle needs to obtain at least the following data: ①The speed of the vehicle to be overtaken in the last time period; ②The acceleration of the vehicle to be overtaken in the last time period; ③The vehicle and the vehicle to be overtaken in the last time period The distance between the vehicle; ④The speed of the vehicle to be overtaken in the last time period; ⑤The acceleration of the vehicle to be overtaken in the last time period; ⑥The distance between the own vehicle and the vehicle to be overtaken in the last time period; ⑦The current vehicle in front Estimated parking distance at vehicle speed; ⑧ Estimated parking distance at current speed of the vehicle.

And output at least the following data: ①The distance between the own vehicle and the preceding vehicle in the current time period; ②The distance between the own vehicle and the overtaken vehicle in the current time period; ③The distance between the own vehicle and the adjacent preceding vehicle in the current time period; ④ The acceleration of the own vehicle in the current time period; ⑤The speed of the own vehicle in the current time period.

The field test SC3 scenario includes the correction of the difference between the horizontal and vertical control strategies adopted by the vehicle when the vehicle in front and the overtaken vehicle are in different relative positions and at different relative speeds, and the correction of the difference in related experience values.

This embodiment also discloses a driver behavior model development device. As shown in FIG. 5, the device includes a memory, a processor, and a program stored in the above-mentioned memory and running on the above-mentioned processor, and the program is processed by the above-mentioned processor. The driver's behavior model development method for autonomous driving as shown in Figure 1 is implemented when the driver is executed.

This embodiment also discloses a computer-readable storage medium, the computer-readable storage medium stores a driver behavior model development program, and when the driver behavior model development program is executed by a processor, the program shown in FIG. 1 is implemented. Driver behavior model development method for autonomous driving.

The above-mentioned embodiments are only preferred embodiments for fully explaining the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or changes made by those skilled in the art on the basis of the present invention are all within the protection scope of the present invention. The protection scope of the present invention is subject to the claims.

Claims

A method for developing a driver behavior model for autonomous driving is characterized in that it includes the following steps:

(1) Create a basic model in the simulation environment based on the transfer relationship between the excitation input and the theoretical output;

(2) Intervene the target vehicle in the field test scene using the same excitation input as the basic model to obtain the actual output data of the target vehicle under the excitation input in the field test scene. The actual output data includes the target vehicle driver's decision data and Operational change data of the target vehicle;

(3) Use the acquired real output data of the target vehicle in the field test scenario to correct the basic model.
The method for developing a driver behavior model for autonomous driving according to claim 1, wherein in step (3), correcting the basic model comprises:

Determine whether the actual output data of the target vehicle in the field test scenario is different from the theoretical output, if so, use the actual output data to update the theoretical output data, and use the updated theoretical output data to modify the basic model.
The method for developing a driver behavior model for automatic driving according to claim 1, wherein the excitation input includes operating change parameters of surrounding vehicles.
The method for developing a driver behavior model for autonomous driving according to claim 3, wherein the incentive input further includes road traffic information parameters and environmental information parameters.
The method for developing a driver behavior model for autonomous driving according to claim 4, wherein the environmental information parameters include one or any combination of illuminance, weather, temperature, humidity, wind direction, and wind speed; the road The traffic information parameters include one or any combination of traffic information parameters on main roads in urban areas, traffic information parameters on suburban roads, national highway traffic information parameters, provincial road traffic information parameters, and high-speed traffic information parameters.
The method for developing a driver behavior model for autonomous driving according to claim 3, wherein the motivation input further includes the driver's age parameter, gender parameter, physiological parameter, psychological parameter, and driving age parameter.
The method for developing a driver behavior model for automatic driving according to claim 1, characterized in that: in step (2), the target vehicle operation change data includes vehicle position change data, speed change data, acceleration change data, and steering angle One or any combination of change data, yaw rate change data, lateral acceleration change data, accelerator opening change data, and brake pedal opening change data.
The method for developing a driver behavior model for automatic driving according to claim 3, wherein the operating change parameters of the surrounding vehicles include vehicle position change data, speed change data, acceleration change data, steering angle change data, One or any combination of yaw rate change data, lateral acceleration change data, accelerator opening change data, and brake pedal opening change data.
A driver behavior model development device, which is characterized in that it includes a memory, a processor, and a program stored in the memory and running on the processor. The program is executed by the processor to achieve The driver behavior model development method for autonomous driving described in any one of 1-8 is required.
A computer-readable storage medium, characterized in that: a driver behavior model development program is stored on the computer-readable storage medium, and when the driver behavior model development program is executed by a processor, any of claims 1-8 is implemented. A described method for developing driver behavior models for autonomous driving.