TU Career Center

mehr Jobs auf derStandard.at

TU Career Center > Abschlussarbeiten

Arbeit "Reinforcement Learning in Behavior Based Embodied Agents", Online seit: 01.09.2009

drucken

		Institut für Computertechnik

Reinforcement Learning in Behavior Based Embodied Agents

Angesprochene Studienrichtungen

Elektrotechnik (Diplom)
Informatik (Diplom)

Beschreibung

Unlike traditional artificial intelligence, embodied agents interact with a – real or simulated – environment. This is done based on the theory, that intelligence is not something acquired through the manipulation of abstracted symbols (like expert systems or theorem solvers) – it rather emerges through the interaction of the agent with the environment. Designing embodied agents helps understanding the dynamics of the relationship between agents and the environment. Embodied agents are equipped with sensors and actuators which are linked through a decision unit and perform various tasks within this environment. In behavior based modeling a task is divided into a sequence of behaviors (e.g. avoid obstacle, navigate, pickup food). Processing this sequence enables the embodied agent to perform the overall task.

Reinforcement learning is a machine learning technique in which agents learn through trial and error. The agent receives a reward or punishment in the form of a scalar value after the execution of a behavior or an action. In this way, its behavior is constantly improved.

Task Description:
In this project, an embodied agent has to be implemented which incorporates the abilities food foraging and homing. They should be realized through reinforcement learning. The embodied agent should be embedded within an existing simulator. Many functionalities, like vision and bump sensors or motion actuators, are already provided there. The simulated environment consists of different obstacles, food sources and other agents.

Skill that can be acquired:
• Profound knowledge in embodied agent modeling
• Established knowledge in reinforcement learning
• Task modeling in behavior based approach
• Object oriented programming in JAVA
• Difference between object oriented and agent oriented programming
• Efficient usage of MASON (Multiagent based simulation toolkit)

Sprachen

deutsch, englisch

Hauptbetreuung

DI(FH) Clemens Muchitsch

+43 1 58801 38479

muchitsch@ict.tuwien.ac.at

Weitere Arbeiten von Institut für Computertechnik

SUCHE

Typ der Arbeit

Institut

Studienrichtung

Studium

Schlagwort

Jobs, Stellenangebote sowie Tipps zur Jobsuche und Bewerbung

im Video bei JobTV24.de