Extra seminar Computer Science, Intelligent Systems - Dr. Kailai Li, Linköping University

Title: Towards Agile Sensing for Autonomous Intelligent Systems

Abstract:

Autonomous intelligent systems are commonly equipped with various sensors of heterogeneous modalities. Based thereon, the fundamental tasks of state estimation and scene perception are performed towards achieving trustworthy autonomy.

The talk revolves around the ongoing enrichment of the concept of agile sensing, considering key criteria ranging from performance and cost-effectiveness to robustness and interpretability. On designing universal inferential frameworks for dynamic state estimation, we present a dedicated guidance on their progressive developments, combining both representational and computational perspectives. The former include discrete-time and continuous-time formulations for system representation, while the latter is categorized into schemes based on nonlinear optimization and recursive filtering. Particularly, we elucidate our explicit considerations on the intrinsic characteristics of uncertain dynamical systems, such as kinematic consistency, probabilistic quantification, and underlying manifold structures.

Throughout these conceptual designs, we conduct real-world validations with system-level developments to provide hands-on insights to agile sensing in mobile robotics. These encompass various sensory modalities, including inertial measurement unit, RGB-(D) cameras, light detection and ranging, ultra-wideband, as well as event cameras. Moreover, some notable progress on efficient scene perception and understanding is presented with a focus on leveraging neuromorphic computing and smart edges.