Reinforcement Learning for Aerial Base Station Placement Problem: Convergence Analysis
- Hauptseminar, Advanced Research Project (MSCSP)
M.Sc. Oleksandr Andryeyev
- Nowadays the use of Aerial Base Stations (ABSs) become more and more attractive approach to increase the system capacity in areas with temporarily high data demand. Instead of centralized placement algorithms, we have developed the self-organized framework. It uses reinforcement learning (RL) to ensure a proper separation distance between ABSs and to mitigate a mutual interference. However, it is still required to analyze the convergence rate of RL in different user distributions and to adapt the algorithm based on these findings.
* Literature study on RL techniques in general and specific to multi-robot deployments;
* Simulative analysis of the Q-Learning convergence rate, its visualization and comparison to other RL techniques;
* Extension/modification of current algorithm to ensure reliable and quick adaptation for different scenarios.
• analytical mind;
• Python knowledge;
• basic knowledge in machine learning.
 O. Andryeyev and A. Mitschele-Thiel, “Efficiency vs. accuracy of aerial base station placement,” in International Conference on Networked Systems 2019 (NetSys 2019), (Garching b. München, Germany), pp. 1–6,IEEE, 03 2019.
 A. S. Tanenbaum and M. Van Steen, Distributed systems: principles and paradigms. Second Edition. Prentice-Hall, 2006
 C. J. Watkins and P. Dayan, “Q-learning,” Machine learning, vol. 8, no. 3-4, pp. 279–292, 1992.