Information-Driven Exploration Strategies for Mobile Robot Olfactory Source Tracking in Confined Environments

This paper presents a comparative analysis of information-driven exploration strategies for autonomous chemical source localization in confined indoor environments using Bayesian particle filters. We implement and experimentally validate two planning algorithms-greedy gradient ascent and stochastic sampling-on a mobile robotic platform equipped with an Alpha MOS NEEM chemical sensor and Gaussian plume dispersion model. Through 18 experimental trials (9 source locations × 2 planners), we achieve localization accuracy of 82% (greedy) and 86% (stochastic) with convergence times of 12.3 and 11.1 minutes respectively. The stochastic planner demonstrates superior environmental coverage (53.1% vs 8.4%) and 11.1% higher success rate in turbulent conditions, revealing a fundamental trade-off between exploration breadth and localization precision.