Neural network modeling of the dynamics of autonomous underwater vehicles for Kalman filtering and improved localization

Autonomous Underwater Vehicles (AUVs) and Remotely Operated Vehicles are used for a variety of underwater operations and deep-sea explorations. One of the major challenges faced by these vehicles is localization i.e., the ability of these vehicles to identify their location with respect to a reference point. The kinematic Extended Kalman filters have been used in localization in a method known as dead reckoning. The accuracy of the localization systems can be improved if a dynamic model is used instead of the kinematic model. The previously derived dynamic model was implemented in real time in UUVSim, a simulation environment. The dynamic model was tested against the kinematic model on various test courses and it was found that the dynamic model was more stable and accurate than the kinematic model. One of the major drawbacks of the dynamic model was that it required the use of numerous coefficients. The process of determining these coefficients was extensive, requiring significant experimentation time. This research explores the use of a Neural Network architecture to replace these dynamic equations. Initial experiments have showed promising results for the Neural Network although modifications will be required before the controller can be made universally applicable.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Industrial Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2020

Author dc:creator

• Balasubramanian, Sharan

Contributors dc:contributor

• Norris, William Robert

• Kalman Filter
• Neural Networks
• Dynamic Modeling
• Autonomous Underwater Vehicle
• RexROV
• EKF

Statement dc:rights

• Copyright 2020 Sharan Balasubramanian

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/108446

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/108446