Events

CS Colloquium: Consistent and Efficient Visual-Inertial State Estimation and Perception

Speaker: Chuchu Chen, University of Delaware

Location: 60 Fifth Avenue, Room 150

Date: Monday, March 17, 2025

As cameras and IMUs are becoming ubiquitous, visual-inertial systems hold great potential in a wide range of applications from extended reality (XR) to autonomous robots. At the core of these systems lies the state estimation algorithm, which aims to estimate the state (i.e., pose) of the sensing platform and the environment by fusing sensor information. Despite significant progress in visual-inertial navigation systems (VINS) and VI-SLAM (Simultaneous Localization and Mapping) over the past decade, critical aspects of system design remain underexplored, limiting their practical deployment. This talk will present advancements in the design of consistent VINS estimation algorithms, where consistency is even more critical than accuracy for ensuring robust and reliable performance. I will also discuss novel techniques for improving estimation efficiency on next-generation edge devices, considering constraints in size, weight, and power (SWaP). Additionally, I will explore the role of multi-sensor fusion and scene understanding in enhancing the reliability of autonomous systems in diverse real-world applications.