Persistent rule-based interactive reinforcement learning

Interactive reinforcement learning has allowed speeding up the process in autonomous agents by including a human trainer providing extra information to agent real-time. Current interactive research been limited real-time interactions that offer relevant user advice current state only. Additionally, provided each interaction is not retained and instead discarded after single-use. In this work, we propose persistent rule-based approach, i.e., method for retaining reusing knowledge, allowing trainers give general more than just state. Our experimental results show substantially improves performance of while reducing number required trainer. Moreover, shows similar impact as state-based advice, but with reduced count.