Sensory predictions of motor commands using LSTM networks (Comparing the sensory predictions quality for a simple RNN versus a LSTM network)

Abstract

This paper investigates whether using an LSTM network to do blind navigation using sensory predictions of motor commands can improve on results shown when less advanced neural networks were used for this task. First three initial experiments are discussed which try to make sensory predictions with a small RNN, then the same experiment is done using an LSTM network. Then the two are compared. Ultimately the LSTM network was not able to perform blind navigation collision free, but it does show some promising results.