A simulation study on reinforcement learning for navigation application

In this paper we have contributed work on implementation of Q-learning, a reinforcement, nonparameter based learning and decision making method. have formulated and demonstrated the Q-learning algorithm via simulation. The work includes formulation of a pseudo-code and development of algorithm taking into account of an application. The application of reinforcement learning is simulated through a conceptual agriculture field, where a robot is commanded to reach at trees and finally delivers the fruits to the storage point (goal). We have studied the effectiveness of g and a. The results show that the learning parameter (g) and the learning rate (a) are the two important parameters to be considered while developing Q-learning based reinforcement algorithm for specific application. We have also established the optimal values of g and a of an application through a simulation study.