"Making Sense : Coupled Sensor Configuration and Path Planning for Autonomous Mobile Vehicles"

Speaker Bio: Raghvendra V. Cowlagi is an Associate Professor in the Aerospace Engineering Department at Worcester Polytechnic Institute, Worcester, MA. Prior to joining WPI in 2013, he obtained a Ph.D. in Aerospace Engineering from Georgia Tech, worked as a postdoctoral researcher at MIT, and worked as a researcher at Aurora Flight Sciences Corp. in Cambridge, MA. His research interests and expertise are in motion planning, optimal control, sensor placement, and estimation for the autonomy of mobile vehicles. His research work is published in leading journal articles and peer-reviewed conference papers, and supported by federal funding agencies including the NSF, the US Air Force, and the US Army. He serves as Associate Editor for the IEEE Transactions on Aerospace & Electronic Systems, the Aerospace Science and Technology journal, and the ASME Journal of Autonomous Vehicles and Systems. He is a member of the IEEE-CSS Conference Editorial Board and the AIAA Guidance, Navigation, and Controls Technical Committee. He is a recipient of the AFOSR Young Investigator Program award, and the AFRL Summer Faculty Fellowship.

Abstract: An autonomous mobile agent wants to move from A to B in an unknown environment. We have the ability to send a few other mobile vehicles (like drones), or query pre-installed fixed assets (like radar sites), to scout the environment before or during the agent’s traversal to help find an optimal plan of motion. What are the best locations to send the scout vehicles, or from which to query fixed assets?

This research problem and its variants can model a wide variety of military and civilian applications including search-and-rescue, natural disaster response, traffic management, and long-range aerial delivery. In such applications, it is desirable to find a near-optimal plan for the mobile agent using a minimal number of scout observations. Finding appropriate locations for the scout vehicles or fixed assets is a problem of optimal sensor placement or sensor configuration, followed by optimal state estimation. Planning the agent’s path in the environment is an optimal path-planning or optimal control problem.

In this talk, I demonstrate that a simultaneous and coupled solution to these problems significantly outperforms traditional decoupled methods where the sensor configuration and estimation are solved before optimal planning/control. I discuss some methods to enable this coupling, based on measures that quantify the expected value of new sensor observations in context to uncertainty information gains (CRIG). Through numerical simulation examples and theoretical analyses, I discuss the current results and future prospects of using such CRIGs in coupled sensor configuration and planning/control problems.