A Human-Robot Collaborative Reinforcement Learning Algorithm
Journal of Intelligent and Robotic Systems
| Hi-index | 0.00 |
This paper introduces the Introspection Approach, a method by which a learning agent employing reinforcement learning can decide when to ask a training agent for instruction. When using our approach, we find that the same number of trainer''s responses produced significantly faster learners than by having the learner ask for aid randomly. Guidance received via our approach is more informative than random guidance. Thus, we can reduce the interaction that the training agent has with the learning agent without reducing the speed with which the learner develops its policy. In fact, by being intelligent about when the learner asks for help, we can even increase the learning speed for the same level of trainer interaction.