Trajectory optimization using dynamic programming and Q-learning

Abstract

This paper presents route optimization in Markov Decision Process (MDP) environment using dynamic programming (DP) and Q-learning, and compares features of two algorithms. Since DP uses recursive function, coding is simple and easy to implement. However, all cases should be calculated to implement DP. Due to this characteristic, the unnecessary calculation rate increases as the model grows for implementation of DP. In this paper, learning is applied to DP for reduction of calculation complexity. Same MDP problem is solved by DP and Q-learning and results of each case are compared and analyzed in this paper.