Speaker Dr. Hyundae Kim
Department of Machanical Engineering, University of Maryland
Date｜Time August 12 (Wednesday), 2026｜15:00
Location 133동 316-1호
Abstract
Uniformly spinning planar LiDAR is often treated as though each completed frame were an instantaneous snapshot of the environment. In practice, however, individual rays are emitted sequentially over a finite scan interval, the completed frame becomes available only after processing latency, and discrete azimuth sampling may miss narrow or moving obstacles between adjacent rays. These temporal and spatial aliasing effects can create a mismatch between the geometry represented by the active LiDAR frame and the environment for which safety must be guaranteed.
This seminar presents a scan-aware framework that converts these sensing effects into explicit and verifiable safety margins. A temporal Minkowski inflation accounts for scan accumulation, processing latency, control-update alignment, bounded frame reuse, perception error, and obstacle motion. A single-scan detectability result combines obstacle feature size, LiDAR angular resolution, visibility range, and within-scan bearing drift to determine when a safety-relevant obstacle must generate a retained LiDAR return. These margins are then incorporated into a robust high-order control barrier function quadratic program with conservative frame switching and a sampled-data inter-update margin. The resulting theorem guarantees a prescribed minimum distance from the true obstacle occupancy on update blocks for which the perception and control certificate conditions are verified. Deterministic certificate-audit simulations separately examine the effects of angular resolution, temporal inflation, positive bearing drift, and conservative single-disc obstacle representations.
Biography
Hyuntae Kim is a Postdoctoral Research Associate at the University of Maryland. He received his B.S. and Ph.D. degrees in Electrical and Computer Engineering from Seoul National University, with the Ph.D. awarded in 2024. Before joining the University of Maryland, he was a Postdoctoral Research Associate in the Department of Engineering Science at the University of Oxford, where he worked on large-scale feedback control and control-oriented modeling for advanced engineering systems. His research interests include robust and learning-augmented control, disturbance-observer-based control, safety-critical and perception-aware control, Gaussian-process-based learning, and runtime assurance with explicit stability and robustness guarantees.
