ScienceAsia 27 (2001) : 177-183 |doi: 10.2306/scienceasia1513-1874.2001.27.177

Genetic Reinforcement Learning with Updating Table of Q-value Function: Obstacle Avoidance Robot

T Leephakpreeda^*, S Limpichotipong and C Netramai

ABSTRACT: This paper presents an alternative approach of the machine learning- the genetic reinforcement learning with the updating table of the Q-value function. The proposed method in updating table is implemented to obtain the reinforcement values of the Q-value function for given state/action pairs corresponding to any policies during exploring environment. To search optimal policies, the fitness of a set of policies for genetic algorithm is defined in terms of the value of the Q-value function. The genetic algorithm and the reinforcement learning are then applied in conjunction to optimize the final control system performance. The effectiveness of the proposed methodology is demonstrated on a real application of the obstacle avoidance robot.

Download PDF

School of Industrial and Mechanical Engineering, Sirindhorn International Institute of Technology, Thammasat University

* Corresponding author, E-mail: thanan@siit.tu.ac.th

Received 31 Mar 2000, Accepted 24 Apr 2001