Progress of the algorithm Q-learning in a field POMDP “application of merger of cubes”

Abstract

This article is part of the field of artificial intelligence (AI), specifically it deals with reinforcement learning techniques to solve a problem modeled by Markov decision processes partially observed (POMDP). The objective in this article is to design an algorithm (based on the algorithm Q-learning) to be implemented on agents, immersed in an environment with partial perceptions and incomplete knowledge, which through their interactions will move together to converge towards a goal. The experimentation of our algorithm on the problem of merger of cubes gave interesting results.