Humboldt-Universität zu Berlin - Mathematisch-Naturwissenschaftliche Fakultät - Institut für Informatik

Die Verteidigung findet digital per Zoom statt. Eine Zoom-Einladung finden Sie hier. (nur mit Informatik-Account)

Abstract

Decision-making is a fundamental component of intelligent behavior, yet it poses significant challenges due to limitations in situational awareness, motor capabilities, and the constraints of time and computational resources. Both humans and other biological agents rely on their ability to anticipate the consequences of their actions - adapting their decisions based on available information about the current and predicted future states of their environment. These principles can also be applied to artificial agents, enabling them to simulate potential outcomes of their actions and select the one that promises the best results.
This thesis explores predictive decision-making mechanisms in artificial agents, focusing on the use of internal simulation to cope with challenges such as uncertainty in perception and action, and environmental complexity. We investigate these principles within the context of RoboCup-SPL, an experimental scenario where humanoid robots autonomously play soccer in teams. This scenario provides a dynamic and uncertain environment in which robots must make rapid decisions, such as determining the optimal direction to kick the ball or coordinating actions with teammates despite limited communication.
Our research demonstrates how internal simulation can enable anticipatory decision-making, allowing robots to choose actions that maximize the likelihood of scoring goals and exhibit complex, emergent collaborative behaviors like passing. The decision mechanism was developed, implemented, and tested both in simulation and on real robots. Additionally, we developed a comprehensive software infrastructure to support these experiments and to analyze the behavior of groups of humanoid robots. The tools and datasets generated through this research provide a robust platform for future studies in robotic decision-making and team coordination.