Automated driving systems are expected to be revolutionary technologies that will reconstruct mobility by improving safety and efficiency. Among the components of automated driving systems, motion planning plays a critical role as it determines how the vehicle reaches its target location from a given origin. The current challenges of motion planning lie in real-time performance and passenger comfort, which is also the main objective of motion planning algorithms. These challenges often occur simultaneously but are treated separately. Neither of these challenges could be solved at a global level of motion planning, which is mainly concerned with route selection and travel time minimization, and could be computed in advance. Consequently, this thesis primarily focuses on local motion planning related to the maneuvering of a vehicle, ensuring real-time performance and passenger comfort.
To address these two challenges, an extended tentacle-based motion planning algorithm is developed. This is an interpolation curve-based algorithm that could work in real-time because there are no optimizers or learning processes that cost a lot of computational resources in the algorithm. The most important factor that affects passenger comfort in a ride is jerk, which is also known as the time derivative of acceleration. The proposed algorithm manages to control the jerk in both lateral and longitudinal directions, thus ensuring ride comfort. Begin with the current state of the vehicle, including velocity, attitude, and steering angle, a series of geometry curves called tentacles is generated and evaluated. The maximum lateral jerk is limited during tentacle generation to avoid excessive impact on passengers during maneuvering.To address these two challenges, an extended tentacle-based motion planning algorithm is developed. This is an interpolation curve-based algorithm that could work in real-time because there are no optimizers or learning processes that cost a lot of computational resources in the algorithm. The most important factor that affects passenger comfort in a ride is jerk, which is also known as the time derivative of acceleration. The proposed algorithm manages to control the jerk in both lateral and longitudinal directions, thus ensuring ride comfort. Begin with the current state of the vehicle, including velocity, attitude, and steering angle, a series of geometry curves called tentacles is generated and evaluated. The maximum lateral jerk is limited during tentacle generation to avoid excessive impact on passengers during maneuvering.
In most motion planning algorithms, geometry path planning and speed planning are treated separately. However, in the proposed planning algorithm, the speed profile is generated based on the selected best tentacle, thus making the speed profile more rational and adaptable to current maneuvers. In addition, the target speed of the speed profile is decided based on road curvature and traffic conditions, ensuring safety and avoiding wasting time on some low-speed traffic participants. More importantly, the speed profile limits the maximum longitudinal jerk, thus making jerk limited in all directions and ensuring passenger comfort.
To evaluate the performance of the proposed motion planning algorithm, simulations using a high-fidelity vehicle model through IPG CarMaker and MATLAB/Simulink are implemented under various conditions. Because this report does not focus on the design of the path-following controller, the vehicle directly uses the output of the planner as control input during simulations. Even so, the algorithm still manages to complete static and dynamic obstacle avoidance as well as adaptive following and overtaking maneuvers at various speed ranges and road conditions. A virtual map of part of the campus of TU Delft is also constructed using Carmaker and used for validation of the proposed algorithm under urban traffic conditions. The proposed algorithm works effectively in complex environments and can operate in real time at a frequency of $20$ $Hz$ while constraining the total jerk.
This research illustrated the capability of the developed tentacle-based motion planning algorithm to ensure passenger comfort and safety under various traffic conditions. Although primarily serving as a concept emphasizing feasibility through simulations rather than immediate on-road verification, the proposed algorithm establishes a foundation for future real-vehicle implementation, thus contributing towards resolving normal driving conditions essential for achieving fully automated driving.