Implement Deep SARSA in Grid World with Changing Obstacles and Testing Against New Environment

Abstract

In this paper, the Deep SARSA method is used to find the path of the robot on 5 × 5 environment with the presence of moving obstacles. This problem is known as Grid world with changing obstacles (GWCO). In GWCO problem, obstacles move on specific paths. Due to a permanent change in the location of obstacles, this can be considered as a dynamic problem. In dynamic problem, the environment is constantly changing. In this paper, we first refer to the applications of Deep learning and Reinforcement learning (RL), then to the details of Grid World and GWCO. Then it is discussed about the Deep SARSA algorithm and the results show that the agent could well find the optimal path and receive the highest reward. After learning the agent, we change the environment and add a new obstacle in the agent’s path. The results show that the agent has been able to quickly propose a new path. Simulation of this paper is done with the Python software and Tensorflow.