A Novel Odor Source Localization Method via a Deep Neural Network-Based Odor Compass

Abstract

Mobile robot-based odor source localization (OSL) has broad applications in industrial and daily-life scenarios. However, subject to the limited sensing capacity of common metal oxide semiconductor (MOS) sensors, the OSL robots still lag far behind their biological counterparts. In this paper, we rethink the odor-source direction estimation paradigm of odor compass and propose a deep neural network (DNN) based method to improve both the accuracy and the generalization ability. The odor compass is composed of four wireless MOS sensors, and a DNN model, which contains a convolutional neural network (CNN) module and a long short-term memory (LSTM) module. An OSL strategy is further designed based on the proposed odor compass. Experimental results validate the feasibility of the proposed method.