Multi-Scenario Constrained Global Motion Planning for Autonomous Vehicles
No Thumbnail Available
URL
Journal Title
Journal ISSN
Volume Title
Sähkötekniikan korkeakoulu |
Master's thesis
Authors
Date
2023-12-11
Department
Major/Subject
Autonomous Systems
Mcode
ELEC3055
Degree programme
Master's Programme in ICT Innovation
Language
en
Pages
56+1
Series
Abstract
This thesis focuses on the trajectory creation process at a startup, called Sensible4, which develops self-driving software for highly autonomous (SAE level 4) route-based vehicles, in particular, low-capacity buses (minivans) for last-mile public transport. Their currently used motion planning methods require a significant amount of manual work in order to create a trajectory for each vehicle. To aggravate this issue, the laboriously obtained trajectory is strongly dependent on the vehicle model, which means that the creation process has to be repeated for every kind of vehicle. Furthermore, it is impossible to perform any post-modification on the trajectory, therefore any mistake made during the creation requires the restart of the whole process. These factors make their trajectory creation process fragile. For this reason, the goal of this thesis is to develop an offline system for global motion planning with multiple scenario constraints, which can overcome the issues related to Sensible4's trajectory creation process. Using motion planning methods, it is possible to implement a system, which is able to create trajectories offline. Given that the trajectory creation is done in a controlled environment considering constraints (such as limits to velocity, acceleration, and jerk), the result can be obtained using less working time and resources. For this thesis, the motion planning system was implemented with the Rapidly-exploring random tree and convex optimization-based algorithms. To test and evaluate the performance of the motion planning system in a standardized manner, the algorithm is tested against several environmental scenarios. These scenarios contain both simple and complex road conditions from the Carla open urban driving simulator's environment, such as multiple lanes and roundabouts. Experiments showed, that the proposed system generated sufficient trajectories in most of the defined driving situations. However, the performance is strongly dependent on the quality of the initial guiding path, which in the current implementation has shown a weakness for multi-lane scenarios.Description
Supervisor
Kyrki, VilleThesis advisor
Abdul Hamid, Umar ZakirKeywords
global motion planning, path planning, speed planning, autonomous vehicles