Deep Reinforcement Learning for Cognitive SONAR Systems

Tag / Zeit: 16.08.2021, 14:40-15:00
Raum: Schubert 4+5
Typ: Regulärer Vortrag
Online-Zugang: Bitte loggen Sie sich ein, damit weitere Inhalte sichtbar werden (bspw. der Zugang zur Onlinesitzung).
Zusammenfassung: In recent years it has been shown that reinforcement learning can also be used for very complex tasks with the help of neural networks. Thereby often strategies are found that exceed results archived by humans. This enables the usage in autonomous tasks in a large number of different areas and applications.One possible application is to control a cognitive SONAR system. The goal is to scan a port area optimally to detect and find incoming objects as fast as possible. Therefore, a simple environment simulation was designed that represents the port area under consideration of random occurring reflections that result in different observations for each scan as well as the use of different novel sonar scanning modes. This work presents an implementation and evaluation of an environment for the training of algorithms to enable the autonomous observation of a port area.