Autonomous navigation of an over-actuated marine platform using reinforcement learning (Master thesis)
The current diploma thesis examines the autonomous navigation of an over-actuated marine platform “Delta Berenike” using reinforcement learning methods. RL aims at finding an optimum route through obstacles in order to identify a specific target. The marine platform is related with certain peculiarities due to hydrodynamic model of the actuators and complex dynamic model,as well as errors from feedback of state variables (current position, orientation and speed) from motion sensors. The RL framework considers navigation problem as a problem of finding the optimal policy of an agent which moves in a stochastic Markov state space. During the agent’s interaction with the environment, reward function plays an important role in the learning process, determining the display format of the state space with the actions. More specifically, reward function is known in advance and is based on the control problem experience. Generally, the problem of the reward function estimation is defined as an inverse reinforcement learning problem. In this study we propose a reinforcement learning framework which controls marine platform and estimates the optimum policy as well as the reward function. The learning process can be implemented as an on line learning algorithm and is focused on linear model in order to describe the value and rewards functions, using a descriptive situation space through appropriate basis function. We have studied the performance of the proposed method using two simulated environments. The environment includes several environmental disturbances such as wind, sea currents and waves. Finally, emphasis was given in the comparison of the method through two known reinforcement learning techniques, the Q-Learning and LSPI algorithm. The results are promising.
|Institution and School/Department of submitter:||Πανεπιστήμιο Ιωαννίνων. Σχολή Θετικών Επιστημών. Τμήμα Μηχανικών Η/Υ & Πληροφορικής|
|Keywords:||Ενισχυτική μάθηση,Ρομποτική θαλάσσια πλατφόρμα,Αντίστροφη ενισχυτική μάθηση,Reinforcement learnings,Robotic marine platform,Inverse reinforcement learning|
|Appears in Collections:||Διατριβές Μεταπτυχιακής Έρευνας (Masters)|
Files in This Item:
|Μ.Ε. ΤΖΙΟΡΤΖΙΩΤΗΣ ΚΩΝΣΤΑΝΤΙΝΟΣ 2016.pdf||1.57 MB||Adobe PDF||View/Open|
Please use this identifier to cite or link to this item:This item is a favorite for 0 people.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.