This thesis presents novel local trajectory planning methods to provide a safe, time optimal, and comfortable passenger ride. Trajectory planning is an essential part of autonomous driving systems, which has been extensively studied for robots during the past few decades. Providing passenger comfort, especially when working within a highly dynamic environment, makes the trajectory planning problem more challenging. Based on the autonomous car situations, three different trajectory planning methods are proposed. The first method is a reactive trajectory planning which works in structured road maps with reference paths. The trajectory speed points are limited based on the road curvature, the traffic rules, and the distances to obstacles. A new algorithm is developed to smooth the speed profile considering the jerk and acceleration constraints. The jerk constraint is defined to provide passengers with a comfortable ride. The acceleration is limited based on the vehicle model and passenger comfort. The vehicle model is determined using the system identification. This approach is suitable for driving in urban areas with dynamic environments in which the obstacle speed changes frequently and the ego car trajectory must react to obstacles while avoiding instant braking or accelerating. In the second trajectory planning approach, a new artificial force vector in three dimensions (longitudinal and lateral position, and speed) allows the autonomous car to follow aspecific path. The vector field is created based on distance from the path and the car speed. It is locally modified by presence of obstacles. By switching between two different vector fields, the vehicle can change a lane or follow another path. The vector field approach is suitable for complicated paths with low traffic such as parking lots. The third trajectory planning approach, Flexible Unit A∗ (FU-A∗), is a new modified tree-based search algorithm in 3-dimensional space (longitudinal and lateral position, and time) in which lane-changing decision is also considered. The energy consumption, time duration, and displacement are integrated in the cost function of the algorithm. This combining of decision-making for lane-changing and following the reference path is one of this thesis’ innovations. The feasibility and reliability of the designed methods are validated through several simulations and implementation on Freie University autonomous cars.
Local Trajectory Planning for Autonomous Driving
Lokale Trajektorieplanung für Autonomes Fahren
2020
ix, 116 Seiten
Miscellaneous
Electronic Resource
Unknown
DDC: | 000 |
Local Trajectory Planning for Autonomous Driving
TIBKAT | 2020
|FOCUSED TRAJECTORY PLANNING FOR AUTONOMOUS ON-ROAD DRIVING
British Library Conference Proceedings | 2013
|