Reinforcement Learning in Noise Enabled Simulated Environment

Résumé

Abstract Reinforcement learning (RL) is a subfield of Artificial intelligence concerned with how agents are trained to become intelligent, so as to take optimum actions in a stochastic environment, in order to maximize the cumulative reward in the long term. The main objective in this field is to find a balance between exploration and exploitation based search strategy. In this context a control problem refers to an objective function of environment state at different points in time and control variables. Optimal control deals with the problem of finding a set of control variable that helps in maximizing the cumulative reward under certain constraints. In case of CartPole, also known as inverted pendulum, the goal here is to move the cart to the left or to the right, so that the pole (pendulum) can stand within a certain angle as long as possible. This work aims to design and develop an intelligent agent for CartPole balancing task using reinforcement learning technique. The actor-critic based reinforcement algorithm was used for training the proposed agent. The results of the proposed agent is compared with a genetic algorithm (GA) based agent. The proposed agent was found to score higher rewards than GA based agent and the training process was significantly faster compared to GA based agent.

Citer ce document

Accès au document

Ce document est hébergé sur son dépôt institutionnel d'origine.

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0