Date of Award

January 2021

Document Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering

First Advisor

Michael Mann


Unmanned Aerial Systems (UASs) has been actively deployed in search and rescue mission,surveillance, and many other applications [1, 2]. To realize UASs fully autonomous, it is essential that Unmanned Aerial Vehicles (UAVs) are equipped with collision avoidance techniques. One of the major challenges for autonomous UAV is that there is a high degree of uncertainty when interacting with moving objects; thus, increasing the likelihood of a collision. This dissertation aims to address these challenges through the following two objectives: 1. Minimize collision scenarios for indoor navigation through a Bayesian network topology. 2. Reduce the costs of path planning in 2D/3D environments for multi-object scenarios. An object's level of uncertainty can be reduced by classifying the object then using a proposed network architecture that is based on the Bayesian probabilistic model topology to determine the possible region of existence within the scene. Object classification finds an object and identifies it using a trained model. The network architecture is then utilized for each detected object using the detected object's type, orientation, and velocity inputs. The output of the network architecture identifies a Safety-Occupied Region (SOR) for the detected objects. The simulation results indicate that the safety occupied region changes with respect to the detected object's state as a function of time. The main contribution of this dissertation is to design a hybrid methodology that takes the proposed network architecture based on the Bayesian network topology to calculate objects' space occupancy binary map and integrates it with 2D/3D path planning algorithms for collision avoidance. This method facilitates alternative collision-free efficient path determinations so the host UAV can reach its destination, allowing it to maneuver safely and closely to other objects. The Hybrid A* algorithm's simulation results show higher efficiency than the PRM algorithm in finding the shortest feasible path in the 2D environment. The proposed multiple 2D combined methodologies based on the Hybrid A* and the 3D A* algorithms are compared to evaluate the effectiveness of collision avoidance with dynamic objects in a 3D environment. Simulation results indicate that the proposed methodologies are effective for collision-free autonomous UAVs.