| Abstract: | Autonomous Underwater Vehicles (AUVs) are playing a larger and larger role in national defense. AUVs equipped with Artificial Intelligence (AI) are capable of numerous useful cognitive tasks, but there are some mission areas where such AUVs could thrive but have little to no role. This thesis proposes the use of AUVs equipped with object detection AI for the defense against swimmer/diver incursions and demonstrates how to do so with a low cost ROV and open-source software. In this thesis an automated search path algorithm is designed which combines PID feedback controls for the depth and yaw with open loop control of the surge motion to convert an ROV, the BlueROV2, into an AUV. An object detection algorithm is implemented using a deep neural network from open-source computer vision libraries to recognize and localize objects from the AUV's onboard camera. These algorithms are combined, and the automated helical search path is tested in both still water and offshore in the ocean. The results of the automated helical search path are compared between the two tests. In particular, the inertial data (acceleration, velocity, and displacement) are compared to understand the influence of the ocean environmental (e.g., waves and currents) on the performance of the control algorithm. Even with the limitation of this kind of kinematic modeling, the closed-loop controls prove to be quite effective at achieving the desired control outcome in the non-ideal ocean environment, especially compared to the open-loop controls. Along the helical path, the object detection was investigated to confirm the validity of the utilized deep neural network underwater. The electronic version of this dissertation is accessible from https://hdl.handle.net/1969.1/198747 |
|---|