Abstract Odor classification is essential in environmental monitoring, gas leak detection, and industrial safety. Although conventional mobile robotic platforms equipped with electronic noses offer advanced gas-sensing capabilities, their performance in confined or cluttered environments is often constrained by rigid structures and limited maneuverability. In this work, we present an olfactory softgrowing robot (oSGR) that integrates bio-inspired, growth-based locomotion with machine-learning (ML) -driven odor classification. Our system comprises a pressurized base enabling contact-free eversion and a custom motorized tip mount housing a multi-sensor array of four metal oxide TGS sensors (2600, 2602, 2611, and 2620) coupled with a passive aspirator for volatile organic compound (VOC) sampling. We provide detailed modeling, design, and structural characterization of the tip mount under multiple actuation configurations, and demonstrate the robot’s olfactory capability through experiments involving four VOCs – ethanol, methane, gin, and acetone. We evaluated two experimental modes: (i) in-transit and static sampling at fixed distances (20, 40, and 80 cm from the source), and (ii) continuous sampling during transit at speeds of 5 cm/s and 10 cm/s. The collected olfactory dataset was used to train twelve widely employed supervised ML classifiers in gas sensing, including k-Nearest Neighbors (kNN), Random Forest, and Linear Discriminant Analysis. The kNN classifier achieved the highest accuracy (99. 88%), demonstrating strong robustness for the olfactory data. Our results highlight the potential of SGRs for contact-free, continuous, in-motion chemical sensing. This unique data acquisition approach reduces detection latency and energy consumption typically associated with conventional stop-and-sense strategies.
Oyejide et al. (Tue,) studied this question.