ECE PhD Prospectus Defense: Hee Jae Kim

ECE PhD Prospectus Defense: Hee Jae Kim

Title: Understanding, Modeling, and Generating Natural Human Motion for Interactive Systems

Presenter: Hee Jae Kim

Advisor: Professor Eshed Ohn-Bar

Chair: Professor Wei-Lun Chao

Committee: Professor Eshed Ohn-Bar, Professor Wenchao Li, Professor Wei-Lun Chao, Professor Hernisa Kacorri, Dr. Ye Yuan

Google Scholar Link: https://scholar.google.com/citations?user=9i7QbK0AAAAJ&hl=en

Abstract: Building safe interactive systems requires the ability to understand human motion, model its uncertainty, and generate plausible future behaviors. Human behavior, however, is shaped by intent, context, and social dynamics that make it inherently stochastic and multimodal. This prospectus studies human motion in safety-critical settings, with the goal of developing models that represent realistic and multimodal behavior in the wild, ultimately equipping systems such as autonomous vehicles and assistive technologies to better anticipate human behavior and respond in ways that are safe, adaptive, and human-centered. Toward this goal, we identify and address core challenges spanning dataset representativeness and generative motion modeling: we leverage simulation for scalable dense social data collection and introduce IMU-based real-world motion data of underrepresented groups such as blind pedestrians to enrich dataset coverage; building on these data, we advance generative modeling through sampling diversification and multimodal probabilistic prediction to capture diverse yet plausible future motions. Preliminary results demonstrate improved diversity and accuracy over competitive baselines, and we further outline future research directions toward characterizing failures in existing autonomous systems when reasoning over the motion of underrepresented groups, with the overarching goal of promoting safe and equitable autonomy for all.