Abstract:
While reward functions are an essential component of many robot learning methods, defining such functions remains a hard problem in many practical applications. For tasks such as grasping, there are no reliable success measures available. Defining reward functions by hand requires extensive task knowledge and often leads to undesired emergent behavior. Instead, we propose to learn the reward function through active learning, querying human expert knowledge for a subset of the agent’s rollouts. We introduce a framework, wherein a traditional learning algorithm interplays with the reward learning component, such that the evolution of the action learner guides the queries of the reward learner. Results of our method on a robot grasping task show that the learned reward function generalizes to a similar task.
Bio:
Christian Daniel is a third year PhD student with Jan Peters at TU Darmstadt’s Intelligent Autonomous System (IAS) lab. His research interests center around hierarchical learning algorithms for autonomous real world robotic systems that can learn from experience in unstructured environments.
Website: http://www.ausy.tu-darmstadt.de/Team/ChristianDaniel
Key Publications:
1) Active Reward Learning http://www.ias.tu-darmstadt.de/uploads/Publications/Daniel_RSS_2014.pdf
2) Learning Concurrent Motor Skills in Versatile Solution Spaces http://www.ias.informatik.tu-darmstadt.de/uploads/Publications/Daniel IROS_2012.pdf
3) Learning Sequential Motor Tasks http://www.ias.informatik.tu-darmstadt.de/uploads/Publications/Daniel ICRA_2013.pdf