الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 136 |  |  |
| | | from | 136 | | |
Abstract
This thesis is concerned with implementation of a path planning algorithm in unstructured environment for autonomous vehicles and introduces a theoretical simulation model taking into account avoiding both static and dynamic multi obstacles. An experimental-based approach has been carried out to validate the theoretical results, High agreement between theoretical and corresponding results have been achieved.
Two different approaches are introduced in this thesis. The first is Path planning approach (I) which utilizes the combination of Gaussian process regression and support vector machines. The regression process predicts the obstacles location for the forthcoming time steps. Moreover, the support vector machines classifier finds the optimal hyperplane with maximum margin to obstacles. Path tracking algorithm is utilized to ensure minimum tracking error, this algorithm is based on a nonlinear model predictive controller considering vehicle dynamics.
The second approach is path planning approach (II), which is based on modified sigmoid function. When obstacles are detected, the proposed algorithm is able to alter the predefined path from global path planning. The algorithm generates a series set of paths to overtake obstacles. Each single path is assessed based on a cost function includes path safety, comfortability, and path length. The parameters of the sigmoid function are optimized using Genetic algorithm for smoothness and minimum path length. Non-linear Model Predictive Controller is utilized for path tracking.
The experimental work utilizing an electric vehicle equipped with multiple sensors; namely Lidar, IMU, camera and GPS. The experimental, beside validating the path planning and tracking algorithm, it ensures the safety and comfortability of the passengers during the car travel. That fulfils the objective of this work.
It is concluded that the simulation results show the effectiveness of the proposed algorithms. The results from approach (I) showed that the maximum lateral acceleration is 0.18g and 0.23g and the maximum lateral error is 0.27 [m] and 0.24 [m] at complicated scenarios. The results from approach (II) demonstrate that lateral acceleration is reduced by 44.27% and 50% furthermore, the lateral error which is significantly reduced by approximately by 91% and 80% after optimization process at complicated scenarios. Comfortability with minimum path length to the target destination have been achieved.