Abstract
In recent years, with the expansion of Internet technology and the improvement of automation levels, self-driving vehicles have received increasing attention from Internet companies and traditional car companies. Self-driving technology is beneficial to society, drivers and pedestrians, and can improve the overall transportation The accident rate has dropped significantly, and the driving mode of autonomous vehicles is more energy-saving and efficient. Therefore, traffic congestion and traffic pollution will be reduced, and it is considered an important means to solve traffic problems in the future. Therefore, it has also attracted the attention of many scholars in the transportation field.
This paper takes the modeling of vehicle lane-changing behavior in the context of autonomous driving as the research direction, establishes an autonomous driving lane-changing trajectory planning model, and conducts simulation analysis of relevant parameters. At the same time, in order to solve the problem of lane change intention generation in the field of autonomous driving, a lane change intention generation model was established to quantify the relationship between the lane change success rate and the lane change intention generation point. Among them, the autonomous driving lane change trajectory planning model is modeled from the perspective of dynamic response of vehicle trajectory planning. It adds a lane change anti-rollover algorithm and a lane change collision avoidance algorithm. It reconstructs the speed change rules of lane change trajectory planning and can achieve The dynamic response to changes in environmental information is a relatively complete autonomous driving lane change trajectory planning model, which can adjust and modify the reaction time and planning step size. And using NGSIM data as environmental information input, four typical lane changing types were simulated. At the same time, the executability of the planned trajectory and various vehicle driving parameters were verified in the CarSim environment; the lane changing success rate model was based on the automatic driving change model. A macroscopic discrete probability model is established based on lane trajectory planning to describe the relationship between the lane change intention generation point and the final lane change success rate, aiming to provide support for the generation of lane change intention for autonomous vehicles.
Chapter 1 Introduction
1.1 Research background and significance
In recent years, with the expansion of Internet technology and the improvement of automation level, autonomous driving technology has received more and more attention. On the one hand, major companies represented by Volvo, Mercedes-Benz, and Tesla Automobile manufacturers are gradually launching technological products such as assisted driving vehicles and semi-automated vehicles. On the other hand, IT technology companies represented by Google and Baidu rely on information technology to find new ways to explore a more intelligent autonomous driving ecosystem, which contains the creativity to subvert the automobile manufacturing industry.
Autonomous driving technology is considered an important means to improve traffic safety and solve traffic congestion problems in the future, and is beneficial to society, drivers and pedestrians. The rapid growth of the market share of autonomous vehicles will steadily reduce the overall traffic accident rate, and the driving mode of autonomous vehicles will be more energy-saving and efficient, so traffic congestion and air pollution will be reduced. Therefore, it has also attracted the attention of many scholars in the transportation field.
However, in recent years, there have been many traffic safety accidents caused by self-driving vehicles. For example, in March 2016, a Google self-driving vehicle had a slight collision with a bus. When the accident occurred, the self-driving vehicle was trying to turn to the right lane. During the lane change, he failed to dynamically respond to real-time environmental information. He mistakenly believed that the bus approaching from the rear would slow down to avoid it, and eventually hit the side of the bus at a low speed. It can be seen from this that although autonomous driving technology develops rapidly, the current lane-changing autonomous driving behavior
still deficiencies, and the safety guarantee is not perfect. Lane-changing behavior is precisely the most basic driving behavior on the highway. Common and highly dangerous driving behaviors have a significant impact on traffic safety and traffic flow characteristics. According to research data from the National Highway Traffic Safety Administration (NHTSA), traffic accidents caused by the lane changing process account for as high as 27% of all statistical traffic accidents [1]. Therefore, in the background architecture of autonomous driving and Internet of Vehicles, modeling research on vehicle lane-changing behavior is a key research content in autonomous driving, including the trade-off of optimal trajectories (convenience, comfort, safety), lane-changing trajectories Issues such as the selection of curves and the traceability of the trajectory. Only by constructing a complete autonomous driving lane changing model can the occurrence of autonomous driving traffic accidents be reduced to the greatest extent and the safety of autonomous driving vehicles can be guaranteed.
1.2 Research objectives and content
1.2.1 Research objectives
: By understanding and studying literature on vehicle lane-changing behavior modeling and analysis and trajectory planning in the field of autonomous driving, we will strengthen our understanding of the microscopic movements of vehicles during the lane-changing process, and become familiar with the current situation of vehicle lane-changing behavior. Stage models and their advantages and disadvantages, and understand the current research status and direction in the field of autonomous driving; establish a dynamic autonomous driving lane change trajectory planning model that can respond to environmental driving information in real time, and can plan appropriate lane changes in different driving environment scenarios Trajectory; establish an autonomous driving lane-changing intention generation model to quantify the relationship between the lane-changing intention generation point and the final lane-changing success rate, aiming to provide support for the generation of autonomous driving lane-changing intentions.
1.2.2 Research content
The research content of this paper mainly focuses on the modeling of vehicle lane changing process in the context of autonomous driving technology. It established its own models for the fields of lane changing decision-making and lane changing trajectory planning, and studied lane changing behavior based on the model.
(1) Established an autonomous driving lane-changing intention generation model: focusing on the forced lane-changing scene on the freeway off-ramp, a discrete macro lane-changing success probability calculation model was established, and the single-lane headway data collected from the video was used to fit the headway. The time interval probability density function is thus used for model calculation, and the accuracy of the model is verified through real vehicle experiments. At the same time, factors such as lane change preparation distance and vehicle speed that affect the success rate of lane change were analyzed.
(2) Established a lane-changing trajectory planning model for autonomous driving: a new dynamic trajectory decision-making algorithm for the lane-changing process was introduced to make the dynamic changes in speed more reasonable. NGSIM data was used as background data for simulation analysis. The simulation results were summarized into four typical lane changing types based on the target lane and speed changes. The impact of planning step size and reaction time on lane changing trajectories was analyzed, and in the CarSim software The executability of the simulation trajectory is verified.
Chapter 2 Current research status at home and abroad The
current research field of lane changing in autonomous driving is mainly divided into lane changing decision research, lane changing trajectory planning research, and lane changing trajectory tracking research. This paper mainly involves the fields of lane changing decision making and lane changing trajectory planning. The following therefore provides a literature review of these two areas.
2.1 Current status of research on lane-changing decision-making
of autonomous driving According to the existing research on lane-changing decision-making of autonomous driving, the contents of lane-changing decision-making mainly include: generation of lane-changing intention, target lane selection, and evaluation of lane-changing conditions [2, 3]. The intention to change lanes occurs when an autonomous vehicle is affected by the speed limit of the vehicle ahead and cannot meet its own driving efficiency, or it needs to enter a ramp due to its inherent driving purpose, and thus the intention to change lanes occurs. Lane-changing condition assessment is when the autonomous vehicle determines the lane-changing condition, and then evaluates the lane-changing conditions to ensure the safety and efficiency of the lane-changing, thereby determining whether the vehicle can change lanes. Selecting a target lane means that after the autonomous vehicle determines that it needs to change lanes, it selects a target lane in the adjacent lane that meets the lane change conditions. According to Rahman et al. [4], existing lane-changing decision-making methods mainly include: rule-based, discrete choice-based, artificial intelligence-based and utility function-based. Here is a brief review of these four methods.
The rule-based autonomous driving lane-changing decision model is represented by Gipps' lane-changing model [5], which means that the vehicle formulates different lane-changing rules for different lane-changing environments, such as minimum safe distance rules, lane-changing obstacle avoidance rules, etc. , the lane-changing vehicle determines whether the current traffic environment meets the conditions for lane-changing based on these rules, and then decides whether to change lanes. Kanaris et al. [6] studied the lane-changing problem of vehicles in the automatic highway system, and proposed a lane-changing condition evaluation method based on the minimum safe distance to evaluate the vehicle lane-changing environment and use the minimum safe distance rule to judge the vehicle. Is it possible to change lanes? Later, when Chen et al. [7] studied the lane-changing decision of autonomous vehicles, they proposed concise and clear lane-changing decision rules and then used multi-attribute decision-making operations, using the analytic hierarchy process and the ranking method that approximated the ideal solution to determine the safety of vehicle lane-changing. and efficiency constraints. The advantage of this decision-making model is that it can transform the driver's subjective and objective awareness into quantitative values, thereby establishing a lane-changing decision model with driver characteristics. Talebpour et al. [8] first proposed a vehicle lane-changing decision model based on game theory in the Internet of Vehicles environment. On this basis, Meng et al. [9] introduced Alireza's structural balance theory and established a rolling time domain control The game theory lane-changing decision model describes the vehicle lane-changing decision as two issues: whether it is worthwhile to change lanes and whether it is safe for the vehicle to change lanes. The game theory is used to comprehensively consider the impact between vehicles, lane-changing safety, and driving efficiency to make a lane-changing decision. Tao decision-making.
2.2 Current status of research on lane-changing trajectory planning
for autonomous driving Although research on autonomous driving has received widespread attention in recent years [32-40], most of them only focus on autonomous driving modeling of car-following behavior [41]. There is insufficient research on trajectory planning of driving vehicles' lane-changing behavior. Existing research mainly focuses on the geometric curve method [42, 43], in addition to the artificial potential field method [44, 45].
In the geometric curve method, according to the different types of geometric curves, it is subdivided into circular trajectory, polynomial trajectory, spiral trajectory, sine and cosine function trajectory, B-spline trajectory and other methods. In addition, some scholars characterize the lane changing trajectory by designing the acceleration curve of the lane changing process, such as forward and reverse trapezoidal acceleration curves. The polynomial method was first proposed by Nelson [46], aiming to construct automatic vehicle guidance trajectories with continuous curvature (AGVs). He proposed Cartesian coordinate polynomial and polar coordinate polynomial trajectories and verified that the polynomial method can improve the trackability of the trajectory. Piazzi and Bianco [47] proposed a trajectory planning method based on fifth-order polynomials. The flatness of the polynomial trajectory curve ensures the executability of the autonomous vehicle trajectory. Different types of geometric linear curve trajectories can be characterized by adjusting parameters. However, in this article, some The parameters are process parameters, do not have actual physical meaning, and cannot be effectively applied to real-time control. Papadimitriou and Tomizuka [48] used a fifth-order polynomial to represent the vehicle lane change trajectory, simplified the obstacles into circles and considered the dynamic constraints during the lane change process, but this strategy can only be applied to obstacles at the beginning of the lane change. , without being able to resolve obstacles that appear during lane changes. Chu et al. [49] studied local trajectory planning in a curved road environment, using a cubic polynomial curve to represent the lane change trajectory, and then calculated the curvature through the trajectory equation, and obtained the vehicle steering angle at each moment to achieve control. In addition, the trajectory Safety, efficiency and comfort during lane changes are also considered in the planning. Shim et al. [50] proposed a sixth-order polynomial trajectory planning method, innovatively introducing the vehicle heading angle and steering angle as boundary conditions to determine the trajectory equation. And use MPC to realize trajectory tracking. In addition, Chen et al. [51] proposed using the quadratic Bezier curve for path planning, and the resulting lane change trajectory has a continuous curvature radius, but this method does not solve the Bezier curve control points in the presence of obstacles. The selection problem does not involve vehicle collision detection. Later, Milam [52] used cubic B-spline curves to replace Bezier curves in trajectory planning. However, when the vehicle behavior changes dynamically, it is still impossible to reasonably determine the number of B-spline curve segments, and the generated lane-changing trajectory cannot be determined reasonably when the vehicle actually changes lanes. The maximum lateral acceleration cannot be effectively controlled during the road. Ren [53] proposed a planning method based on the trapezoidal acceleration curve, derived the yaw rate and yaw angle acceleration, and then designed a yaw rate tracker to achieve lane changes by applying non-modal sliding mode technology, but only assumed The lateral acceleration satisfies the forward and reverse trapezoidal curves, but the longitudinal speed cannot be adjusted in real time. In addition, some scholars have conducted comparative studies on different geometric curves used in lane changing trajectories. Chee and Tomizuka [54, 55] comparatively studied four different desired trajectories, polynomial trajectory, circular trajectory, trapezoidal acceleration trajectory, cosine function trajectory and two trajectory tracking algorithms and selected the trapezoidal acceleration trajectory as the virtual desired trajectory, and gave A sliding mode controller algorithm was developed to improve the stability of the system. However, they studied the lane-changing vehicle in isolation, assuming that there were no other surrounding vehicles that affected the lane-changing process. Sledge and Marshek [56] compared several commonly used lane change trajectory candidate curves, used the maximum speed as an additional discriminant index and transformed it into an optimization problem under boundary condition constraints to solve. Zhang et al. [57] used sinusoidal curves, positive and negative trapezoidal acceleration curves, and spirals to express lane changing trajectories, and used initial point, end point states, and other constraints (such as collision avoidance) to determine the range of parameters, and finally based on optimization The target determines the value of the parameter. Later, Zhang et al. [58] added a cost function that considers driving comfort and efficiency to perform trajectory optimization, and used a cubic polynomial equation to characterize the lane-changing geometric curve. 55] compared and studied four different desired trajectories, polynomial trajectory, circular trajectory, trapezoidal acceleration trajectory, cosine function trajectory and two trajectory tracking algorithms, and selected the trapezoidal acceleration trajectory as the virtual desired trajectory, and gave a sliding mode controller algorithm Thereby improving the stability of the system, however, they studied the lane-changing vehicle in isolation, assuming that no other surrounding vehicles would affect the lane-changing process. Sledge and Marshek [56] compared several commonly used lane change trajectory candidate curves, used the maximum speed as an additional discriminant index and transformed it into an optimization problem under boundary condition constraints to solve. Zhang et al. [57] used sinusoidal curves, positive and negative trapezoidal acceleration curves, and spirals to express lane changing trajectories, and used initial point, end point states, and other constraints (such as collision avoidance) to determine the range of parameters, and finally based on optimization The target determines the value of the parameter. Later, Zhang et al. [58] added a cost function that considers driving comfort and efficiency to perform trajectory optimization, and used a cubic polynomial equation to characterize the lane-changing geometric curve. 55] compared and studied four different desired trajectories, polynomial trajectory, circular trajectory, trapezoidal acceleration trajectory, cosine function trajectory and two trajectory tracking algorithms, and selected the trapezoidal acceleration trajectory as the virtual desired trajectory, and gave a sliding mode controller algorithm Thereby improving the stability of the system, however, they studied the lane-changing vehicle in isolation, assuming that no other surrounding vehicles would affect the lane-changing process. Sledge and Marshek [56] compared several commonly used lane change trajectory candidate curves, used the maximum speed as an additional discriminant index and transformed it into an optimization problem under boundary condition constraints to solve. Zhang et al. [57] used sinusoidal curves, positive and negative trapezoidal acceleration curves, and spirals to express lane changing trajectories, and used initial point, end point states, and other constraints (such as collision avoidance) to determine the range of parameters, and finally based on optimization The target determines the value of the parameter. Later, Zhang et al. [58] added a cost function that considers driving comfort and efficiency to perform trajectory optimization, and used a cubic polynomial equation to characterize the lane-changing geometric curve.
2.3 Current research status of autonomous driving lane change tracking
Based on the existing research on lane-changing trajectory tracking control of autonomous driving, trajectory tracking control is divided into sliding mode control, inversion algorithm, adaptive, PID controller, fuzzy logic, model predictive controller and other methods. The following is a brief review of sliding mode control and inversion algorithms.
Sliding mode control is a highly robust method for controlling nonlinear systems. Applying sliding mode control theory to lane-changing trajectory tracking control of autonomous vehicles has the advantages of fast control response, insensitivity to disturbances, and easy operation. . Hatipoglu et al. [62] established a two-layer architecture lane-changing controller for autonomous driving, converting the lane-changing control problem into an equivalent reference trajectory tracking control problem. Later, Luo et al. [60] considered the deviation of position coordinates and heading angle between the actual trajectory and the reference trajectory based on the dynamics of inter-vehicle communication, and applied the sliding mode controller theory to correct the deviation, thereby achieving vehicle lane-changing trajectory tracking control. , but the requirements of comfort and driver characteristics are not considered in the trajectory tracking control process.
The backstepping algorithm decomposes the complex linear system into multiple subsystems, and then analyzes each subsystem. The backstepping algorithm is used in the lane-changing trajectory tracking control of autonomous vehicles to ensure the stability of the vehicle lane-changing control system. sex. Rong et al. [63] studied vehicle lane changing or overtaking tracking control and established the pose error between the current and desired poses. The pose includes the vehicle's transverse and longitudinal position and the vehicle's yaw angle, and used an integral inversion algorithm to realize vehicle lane changing. Trajectory tracking control. Subsequently, Guo et al. [64] also introduced the integral inversion algorithm in the study of trajectory tracking control, constructed a new pose error variable, and derived a vehicle lane-changing trajectory tracking controller with a closed-loop control structure. You et al. [65] designed a trajectory tracking controller through the inversion algorithm. This controller can ensure the global convergence of the lane change project and transform the trajectory tracking problem into a determined bounded control input. Controllers based on the inversion algorithm all pass The Lyapunov function proves the stability and global convergence ability of the tracking controller.
This article mainly focuses on the lane-changing intention generation part and the lane-changing trajectory planning part during the lane-changing process of autonomous vehicles. Trajectory tracking research has not been carried out.
Chapter 3 Lane-changing intention generation model based on lane-changing success probability
3.1 Introduction
As an important basic component of the highway, the exit ramp diversion area has always been a bottleneck area of road traffic capacity, and therefore is also an area where traffic accidents occur frequently. Vehicles exiting the ramp in this area need to go through processes such as lane changing, slowing down, and entering the ramp, which causes the vehicles to be redistributed among the lanes, making the traffic flow chaotic and complex, and posing a great potential traffic safety hazard. Judging from the statistics of traffic accidents on expressways in operation in our country, the accident rate in the off-ramp diversion area of the expressway is significantly higher than the basic road section. Therefore, it is necessary to conduct modeling research on the lane-changing behavior of vehicles with the intention of off-ramp on the highway to improve driving safety in the highway environment.
In the existing field of autonomous driving research, the research on lane changing behavior is mainly divided into three parts: lane changing decision-making, lane changing trajectory planning and lane changing trajectory tracking. The field of lane changing decision-making mainly includes rule-based methods, artificial intelligence algorithm-based methods and utility function-based methods. The field of lane changing trajectory planning mainly includes geometric curve method, search method and model predictive control method. The field of lane change trajectory tracking includes methods such as sliding mode control and inversion algorithms. However, in the current research, most of the lane-changing decision-making content only focuses on the free lane-changing behavior of vehicles in a straight highway section. The reason for the lane-changing intention at this time is often that the expected driving speed is greater than the current lane driving speed. Therefore, The focus is on whether the vehicle intends to change lanes. However, in the expressway off-ramp scenario, whether the lane change intention is generated has nothing to do with the speed demand, but is determined by the willingness to exit the ramp. The timing of the lane change intention is related to the vehicle's lane change success rate, so attention should be focused When the intention to change lanes occurs. This model attempts to fill this gap, focusing on the generation of lane-changing intentions and judgment of lane-changing timing in the field of lane-changing decision-making. Since this model is biased toward a macroscopic traffic flow model, the lane change trajectory planning module is simplified and the global trajectory planning method is used to determine the latest lane change point, and the lane change trajectory tracking module is not involved for the time being. The research object of this model is forced lane changing behavior in highway off-ramp scenarios. In this scenario, the closer the vehicle lane-changing intention generation point is to the ramp exit, the smaller the success rate of lane-changing to the ramp exit. At the same time, the lane-changing behavior is also affected by the traffic flow environment. This article hopes to establish a certain universality of lane changing probability model in highway off-ramp scenarios to describe the functional relationship between the position of the lane changing intention generation point and the lane changing success rate under different traffic flow states, so as to obtain information that can guide autonomous driving vehicles. The theoretical system of lane-changing decision-making enables autonomous driving lane-changing vehicles to judge and determine the location of the optimal lane-changing intention generation point based on the headway distribution and its own vehicle speed in the current environment, and find the appropriate lane-changing opportunity to change. On the one hand, the position of the lane-changing intention generation point can be appropriately adjusted to ensure a higher probability of lane-changing success, thereby satisfying the vehicle's lane-changing (entering the ramp) intention. On the other hand, it enables off-ramp vehicles to switch to low-speed lanes in time, thereby reducing the impact of on-ramp traffic on main road traffic and improving the traffic capacity of highway sections. This model is a macro prediction model, so it does not involve the microscopic game behavior between vehicles, but uses traffic environment information to affect lane-changing vehicles in the form of lane headway distribution.
3.2 Model establishment
The scenario described by this model is shown in Figure 3-1, which is the most basic forced lane change scenario on a freeway off-ramp. In order to enter the ramp exit deceleration lane at the speed limit, the vehicle SV must start at or before point C. Change lanes from high speed lane to low speed lane. Point B is the latest lane change starting point, LB is the latest lane change trajectory of three lanes, LF is the latest lane change trajectory of two lanes, LD and LG are the actual lane change trajectories. In order to determine the remaining lane-changing preparation time T of the lane-changing vehicle SV, it is first necessary to perform global optimization of the lane-changing trajectory, that is, to determine the position of point B. The trajectory planning method here will be introduced in detail in Chapter 4. Assuming that point A is the vehicle's lane-changing intention generation point, the vehicle SV starts to search for a suitable headway at point A. The core of this model is to establish the functional relationship between the location of the lane change intention generation point and the lane change success rate.
Figure 3-1 Schematic diagram of forced lane change in highway off-ramp scene