A comparative study of learning based control policies and conventional controllers on 2D bi-rotor platform with tail assistance

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Fen Bilimleri Enstitüsü, Türkiye

Tezin Onay Tarihi: 2019

Tezin Dili: İngilizce

Öğrenci: HALİL İBRAHİM UĞURLU

Asıl Danışman (Eş Danışmanlı Tezler İçin): Afşar Saranlı

Eş Danışman: Sinan Kalkan

Özet:

With the developing technology, multi-rotor platforms have become widespread and their control has become an important problem. In this thesis, we analyze physical extensions and control approaches for better control of rotor platforms. The first main contribution of the thesis is whether a tail-appendage that is attached under a multi-rotor platform can improve the multi-rotor's performance. Moreover, we used conventional control approaches as well as Deep Reinforcement Learning to learn a policy for controlling rotor platforms with or without tail appendage. To obtain better training and testing performance with Deep Reinforcement Learning, we adopted a curricular learning approach, where the difficulty of training samples is gradually increased. For the experiments, a two-dimensional simulation environment is developed to simulate a bi-rotor flying system, the counterpart of quad-rotors in three-dimensions. Both control strategies are rigorously analyzed for controlling the platform with and without tail appendage in this simulation environment.