BDBComp
Parceria:
SBC
Aprendizagem da Navegação em Robôs Móveis a Partir de Mapas Obtidos Autonomamente

Sildomar T. MonteiroCarlos H. C. Ribeiro

We analyzed the performance of reinforcement learning algorithms in a navigation problem where maps were generated autonomously by a mobile Magellan Pro robot. The algorithms assessed in this study were Q-learning, Sarsa and Q(), and a method to build variable resolution cognitive maps of the environment was implemented in order to create a performance verifier of the learning algorithms. The learning algorithms presented satisfactory performance, although with a graceful degradation of efficiency due to state ambiguity. The Q-learning algorithm accomplished the best performance over the experiments, followed by the Sarsa algorithm. The Q() algorithm had its performance restrained by experiments parameters. The cognitive map learning method revealed to be efficient and allowed adequate algorithms assessment. Clique no link abaixo para buscar o texto completo deste trabalho na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato: bdbcomp@lbd.dcc.ufmg.br
     Mantida por:
LBD