Gas Source Localization of possible dangerous chemical substances by combining a dispersion map using Structural Similarity Index

In the last years, olfaction has been a challenging sensing modality for intelligent systems, now it is possible to detect and recognize chemical substances for a variety of applications. An existing application is to equip mobile robots with electronic noses for odor based location or track. Furthermore, from the conceptual point of view, the use of smell alone cannot detect and localize physical objects, thus to find and detect an odor source, other sensors and techniques are required than can provide special mapping features. This paper presents a novel effective portable system to generate a dispersion map to identify possible dangerous chemical substances using three agent robots with an e-nose circuit, seven ultrasonic, and two infrared sensors used for the obstacle detection system. The proposed e-nose circuit uses an array of three metal oxide sensors (MOX) within a sealed aluminum box and a small fan to inhale air, dope the sensors and clean them. The entire data acquisition and motor control run on Teensy 3.6, while for the direct and inverse kinematic solution a Raspberry Pi 3 was used, all integrated into each of the three agent robots to determine the position using the cinematic model. The system performance is tested with substances of interest such as alcohol, TNT, toluene in a controlled environment. Measurements showed that using a principal component analysis over the odor, and position data is possible to generate a dispersion map. However, after applying the Structural Similarity Index, the visualization of the dispersion of the substances of interest on the map of the sample room is notably enhanced considering the different selectivity of sensors and mixed vapor effects in the environment.