Deep reinforcement learning for autonomous quadcopter guidance

Date

2023-7-5

Author

Aydınlı, Şevket Utku

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

111
views

141
downloads

Cite This

This thesis examines the quadcopter guidance problem and proposes deep reinforcement learning-based approaches for the solution of this problem. Within the scope of this thesis, the quadcopter guidance problem has been examined from various aspects such as guidance in 2D and 3D spaces, simultaneous arrival and impact time control. The suitability of the proposed approaches for quadcopter guidance problem under non-ideal conditions such as target movement, measurement delay and system delay has been evaluated. Firstly, deep reinforcement learning-based guidance approaches were compared with classical closed-form guidance algorithms and model-based computational guidance algorithm in both simulations and experiments for stationary and moving targets. Secondly, the effect of non-ideal conditions such as target movement, system delay, and measurement delay on the performances of the proposed approaches has been examined. Thirdly, a novel three-stage learning-based approach was proposed for the solution of the simultaneous arrival and impact time control problems. Finally, the usability of deep reinforcement learning-based approaches to solve the quadcopter guidance problem in 3D was investigated. The simulation and experimental results show that the proposed approaches are suitable solutions for the above-mentioned quadcopter guidance problems.

Subject Keywords

Deep reinforcement learning, Guidance algorithm, Proportional navigation, Model predictive control

URI

https://hdl.handle.net/11511/104773

Collections

Graduate School of Natural and Applied Sciences, Thesis