Smart Chemical Sensing Payload for Emergency Response Uncrewed Autonomous Systems

Abstract A novel smart-sensor payload for uncrewed autonomous systems and emergency responders that automatically detects, estimates, and locates chemical sources is presented. The smart-sensing device fuses Bayesian inference machine learning with information-theoretic motion planning for fast source estimation and localization. More specifically, chemical concentration is measured by a newly developed microelectromechanical-system (MEMS)-based sensor, the location and size of a chemical leak are estimated by a Bayesian inference machine learning process, and information-theoretic motion planning is used to optimally guide the user or an autonomous mobile robot during the search process to improve the speed and accuracy of localizing and quantifying a leaking gas source. Experiments are performed that compare the device’s performance under two different motion planning methods: (1) moving the device as instructed by the information-based, guided motion planner and (2) randomly moving the device for search (baseline approach). By following the device’s visual cues on where to take measurements (guided motion method), on average, the smart chemical sensor locates a source over 170% faster than moving the sensor randomly (baseline unguided motion method). Additionally, the leak localization error is less than 6.4% (0.325 m). Finally, live methane gas release experiments are performed to further demonstrate the real-world application of the smart handheld chemical sensing device.