"Reinforcement Learning for Several Environments: Theory and Applications", Andreas Matt and Georg Regensburger

Until now reinforcement learning has been applied to learn the optimal behavior for a single environment. The main idea of our approach is to extend reinforcement learning to learn a good policy for several environments simultaneously.

We link the theory of Markov decision processes (MDP) with notions and algorithms from reinforcement learning to model this idea and to develop solution methods. We do not only focus on rigorous models and mathematical analysis, but also on applications ranging from exact computations to real world problems.

Details can be found in the publications section and in the joint PhD thesis of Andreas Matt and Georg Regensburger "Reinforcement Learning for Several Environments: Theory and Applications"

Construction of the Che-bot Roboter, Departamento de Computacion, University of Buenos Aires, Argentina
The robot soccer team is currently working at the Che-Bot, a robot for robot soccer games and scientific research. Read more.

Online reinforcement learning for several environments, Diego Bendersky
Target following is taught by two different reinforcement functions, keep angle and keep distance, which can be learned simultaneously with the general policy iteration algorithm. A neural network implementation is given and real robo experiments are conducted.

Reinforcement Learning problems with large action sets, Flavia Paganelli, Patricio Otamendi
Macro-action techniques and time variable action sets are tested to accelerate learning in problems with a large amount of different actions for each state.

Neuro Q-Learning application in mathematics of finance and economy, Alex Weissensteiner
Making optimal consumption - investment decisions in a life cycle approach (with a finite horizon) is a highly complex task. In this work Neuro Q-Learning is used to learn optimal decision rules. Read more.

We offer Bachelor and Master Students to participate in our project. Please contact us - Andreas.Matt@uibk.ac.at.