Session: 20-17-01: Rising Stars of Mechanical Engineering

Paper Number: 172852

Long-Range Signal Detection in Cluttered Environments With Dynamic Non-Convex Signal Fields 

Long-range signal source detection in cluttered environments presents a significant challenge, mainly due to the detrimental effects of the environment on signals propagation paths, sources motion, and presence of extraneous signal sources. Regardless of the quantity or sensitivity of the sensors used, reliable sensing in these environments remains unattainable due to strong influence of the environment on signals propagation paths leading to multi-path signals, signal distortion, and degradation.  One affordable solution involves using low-cost autonomous unmanned vehicles (AUVs) that can leverage their mobility and actively explore the environment using an adaptive sensing and navigation framework known as Extremum seeking control (ESC). ESC is a unique model-free control theoretical approach that can tackle continuous-time optimization problems, including open-world dynamic systems problems with physical domains. This quality can be leveraged to use ESC for signal source seeking problems which involve steering an AUV (aka the seeker) to sources of signals in an a priori unknown signal map (or field), without knowledge of its position or the position of the sources. However, existing ESC approaches suffer from a lack of rigorous experimental validations and are predominantly restricted to theoretical scenarios featuring convex maps or maps with mild non-convexity. Moreover, these methods inadequately tackle the complexities inherent in real-life applications, including time and energy constraints. This project will advance ESC theory and applications by formulating methods to: (1) improve the convergence speed in strictly convex multi-dimensional maps; (2) identify locally concave subspaces of the domain and prevent divergence in those spaces; (3) escape local or undesired extrema in highly non-convex maps; (4) enable effective signal isolation, robust tracking, and adequate coverage in domains with multiple moving sources; and (5) use sensor motion to perturb the estimates of the derivatives and save power and time; and (6) guarantee safe navigation in dynamic cluttered environments. The established continuous-time optimizers and eigenvalue decomposition method can be adapted for continuous-time training of neural networks and reduced-order controls, respectively. The innovations will benefit applications requiring signal source localization under challenging conditions, prevalent in critical sectors of national security, environmental monitoring, and industry. The application scenarios are characterized by a priori unknown, highly nonconvex signal maps (created by clutter or extraneous sources), deprivation of position information, and limited sensors. Examples include: identifying abnormal equipment sounds in cluttered industrial settings or detecting leaks in cluttered space stations for predictive maintenance; tracking and studying marine animals using their acoustic signatures; and locating signals from disaster victims in search and rescue (SAR) operations.

Presenting Author: Zahra Nili Ahmadabadi San Diego State University

Presenting Author Biography: Zahra Nili Ahmadabadi is an assistant professor in the Mechanical Engineering Department at San Diego State University (SDSU). Nili is currently the associate editor of Sage Journal of Vibration and Control. Her current research interests include robot perception and control, and nonlinear systems.

Authors:

Zahra Nili Ahmadabadi San Diego State University