Challenges for Relational Reinforcement Learning
Martijn van Otterlo, Kristian Kersting · University of Twente Research Information · 2004
We present a perspective and challenges for Relational Reinforcement Learning (RRL). We first survey existing work and distinguish a number of main directions. We then highlight some research problems that are intrinsically involved in abstracting over relational Markov Decision Processes. These are the challenges of RRL. In addition, we describe a number of issues that will be important for further research into RRL. These are the challenges for RRL and deal with newly arising issues because of relational abstraction.