A new algorithm to track dynamic goal position in Q-learning
Soumishila Mitra, Dhrubojyoti Banerjee, Amit Konar, R. Janarthanan · 2012
In this paper we have evaluated a new approach of Q-learning based on knowledge update in more extended environment. After learning at a fixed goal position, it is convenient for a robot to reach to the fixed destination from where it has started learning. With the new approach we can change the destination even after learning. The above process is evaluated with the concept of state-action pair values. The implemented idea focuses on the fact that only one time learning is required after reaching the first destination. This new application in Q-learning greatly improves the time-management by reducing the frequency of learning.