Paper title: LEARNING ONLINE SPATIAL EXPLORATION BY OPTIMIZING ARTIFICIAL NEURAL NETWORKS ASSISTED BY A PHEROMONE MAP
Author(s): BOGDAN-FLORIN FLOREA, OVIDIU GRIGORE, MIHAI DATCU,
Abstract: This paper addresses the problem of online spatial exploration by using reflex agents controlled by neural networks (multilayer
perceptrons) optimized using a form of genetic algorithms in combination with a pheromone map that acts as information
storage and exchange medium. We have used a fitness function which also contains information about the structure of the
problem and progressively changes with the number of generations, ranging from an emphasis on basic behavior related to
obstacle avoidance and moving towards the exploration frontier in the early generations to an emphasis on the exploration
performance as the number of generations’ increases. We have shown that the reflex agents optimized using the technique
proposed in this paper are capable to solve the exploration problem even with a small number of neurons.
Keywords: Autonomous agents, Cooperative systems, Genetic algorithms, Intelligent robots, Mobile agents Year: 2017 | Tome: 62 | Issue: 2 | Pp.: 209-214
Full text : PDF (252 KB) |