The ultimate goal of self-driving technologies is to offer a safe and human-like driving experience. As one of the most important enabling functionalities, trajectory planning has been extensively studied from the perspective of safety. However, human-like trajectory planning on curved roads has rarely been studied. In this paper, we characterize and model human driving using extensive experimental driving collected on an urban curved road with 30 participants (10 experienced and 20 novice drivers) and five vehicles of different types. Differential global positioning system (GPS) is used to measure vehicle positions in high precision. We study factors that affect the driving trajectory, including vehicle speed, road curvature, and sight distance. We find that the human drivers typically do not follow lane centerline and the human-driven trajectories are very different from planners like rapidly exploring random tree (RRT). To generate human-like driving trajectory, we develop a data-driven trajectory model using general regression neural network (GRNN). The model was validated in various cases with promising performance.

In the coming decades, it is a universal consensus that autonomous vehicles (AVs) and human driven vehicles will share the traffic roads. Trajectory planning of AVs has been extensively studied from the perspective of driving safety and riding comfort. However, human-like trajectory planning has rarely been studied. In this paper, we characterize and model human driving trajectories using real vehicle field test data collected on 5 two-way & two-lane urban curved roads with 20 experienced drivers and 3 experimental vehicle. A differential global positioning system (GPS) and a inertial navigation system (INS) are used to measure vehicle positions and velocities in high precision. We study the trajectory characteristics of experienced drivers on curved two-lane roads, especially the relationships between the vehicle trajectories on bidirectional two lanes. Based on long short-term memory neural network (LSTM NN), we develop a data-driven trajectory model to generate human-like driving trajectories. By comparing with other three modeling methods, the LSTM NN model was validated and tested in various cases with promising performance.

To make intelligent vehicles obtain human drivers' steering characteristic, a new single point preview-based human-like driver model is proposed, which contains a preview decision module and a steering wheel angle calculation module. Enlightened by the visual gaze mechanism of human drivers when passing a curved road, the preview decision module is established using Takagi-Sugeno fuzzy inference system (T-S FIS) to adaptively adjust the preview point position in both longitudinal and lateral direction, and the steering angle calculation module would use the adjusted preview point to generate steering command based on pure pursuit algorithm. Ant colony optimization (ACO) method is used to optimize the fuzzy rules in the preview decision module according to the similarity of trajectories between the proposed driver model and human drivers. The proposed human-like driver model is verified on a two-lane urban curved road. Five experienced human drivers' driving trajectories under different speeds are collected for the verification. After the preview decision module optimization done in the PreScan/Simulink simulation platform, the proposed human-like driver model shows higher similarity with experienced human drivers than the driver model with fixed preview distance or the driver model only with changeable longitudinal preview distance.

This paper proposes a dynamic path planning method based on discrete optimization applied to suburban highways or expressways. We optimize the generated candidate points according to the cost function and then generates a local path through the quintic spline curve. The kinematics and obstacle boundary conditions are set up to improve the reliability of the planned path. Meanwhile, the two-dimensional normal distribution obstacle cost function, comfort cost function and acceleration cost function are designed to evaluate candidate points and speed. Various types of roads scenes, including dynamic and static obstacles, such as straight roads, S- bend, are established to verify the method’s feasibility. The simulation shows that the method can efficiently avoid various obstacles and plan an ideal path that complies with traffic laws.