Olfactory telerobotics. A feasible solution for teleoperated localization of gas sources?

摘要

Olfactory telerobotics consists in augmenting the sensing capabilities of a conventional teleoperated mobile-robot to acquire information about the surrounding air (i.e. smell, wind-speed, etc.) in addition to the usual audio and video streams. Conceptually, this allows for new and improved applications, among which the most relevant are those related to gas-source localization (GSL). That is, searching through telerobotics for one or multiple gas-emission sources, such as hazardous gas-leaks in industrial facilities or the CO2 signature of trapped survivors in collapsed buildings. Notwithstanding, both the needed sensing-technology for the robot as well as the olfactory feedback-interfaces for the human operator are relatively recent, and might still not meet all the requirements of such applications. This work is therefore meant to assess the current feasibility of olfactory telerobotics to address real-world GSL problems, and accordingly, to determine which aspects play the most important role for its success, or otherwise, might be constraining its usefulness. We have collected to this end a dataset composed of 60 experiments where volunteer operators had to locate and identify hidden gas-source among several identical candidates with an olfaction-enabled robot and under realistic environmental conditions (i.e. uncontrolled and natural gas-distributions). We analyse this data to determine the overall search accuracy and intuitiveness of the system, considering that none of the operators had any prior experience with it, and study the importance of the employed sensory-feedback and how they were employed during the experiments. We finally report different findings, from which we highlight that the tested telerobotics system allowed the operators to correctly identify the source in 3 out of 4 attempts, and that the underlying human search-strategy appears to be a probabilistic-driven behaviour that favours semantic and visual information over the robot's gas and wind measurements. (C) 2018 Elsevier B.V. All rights reserved.