An Efficient Learning Based Autonomous Exploration Algorithm For Mobile Robots*

Abstract

In this paper, a novel autonomous exploration algorithm is proposed to achieve efficient exploration task of an unknown environment in terms of the shortest path. First, a new neural network based on the variational autoencoder, LMPnet, is proposed to predict a series of local maps with projected obstacles of unknown areas. Then, a deep Q-network with long-short term memory (LSTM) structure, ETPNet, is proposed to generate piecewise local target points based on the predicted local maps where the reward function is designed to favor shorter length of the local path and larger information gain. Experimental results demonstrate that the proposed algorithm achieves good performance in reducing exploration time.