Agent-advising Approaches in an Interactive Reinforcement Learning Scenario

Reinforcement learning has become one of the fundamental topics in thefield of robotics and machine learning. In this paper, we expand theclassical reinforcement learning framework by the idea of external interaction to support the learning process. To this end, we review anumber of proposed advising approaches for interactive reinforcement learning and discuss their implications, namely, probabilistic advising,early advising, importance advising, and mistake correcting. Moreover, weimplement the advice strategies for interactive reinforcement learning based on a simulated robotic scenario of a domestic cleaning task. Theobtained results show that the mistake correcting approach outperforms apurely probabilistic advice approach as well as the early and importance advising approaches allowing to collect more reward and also to converge faster.

@InProceedings{CWMFW17, 
 	 author =  {Cruz, Francisco and Wueppen, Peter and Magg, Sven and Fazrie, Alvin and Wermter, Stefan},  
 	 title = {Agent-advising Approaches in an Interactive Reinforcement Learning Scenario}, 
 	 booktitle = {IEEE International Conference on Development and Learning and Epigenetic Robotics (ICDL)},
 	 editors = {},
 	 number = {},
 	 volume = {},
 	 pages = {209--214},
 	 year = {2017},
 	 month = {Sep},
 	 publisher = {},
 	 doi = {10.1109/DEVLRN.2017.8329809}, 
 }