Student Seminar by Mr. Zian Wang Peter on Mar 11, 2026, 1PM

Title: Toward Unified Risk Field Modeling for Interactive Autonomous Driving: A Comparative Study and Future Directions

Speaker: Mr. Zian Wang Peter (Department of Data and Systems Engineering)

Date: Mar 11, 2026 (Wednesday)

Time: 1:00 pm - 2:00 pm

Venue: Room 8-28, Haking Wong Building, The University of Hong Kong

Registration Link: https://forms.office.com/r/WssbpUgZaj

ITS Student Committee will provide a beverage for registered participants.

Abstract: Quantifying interaction risk in mixed traffic remains a fundamental challenge for autonomous driving systems. Traditional approaches rely on low-dimensional metrics such as time-to-collision (TTC), which struggle to capture the spatially distributed and temporally evolving nature of hazards in complex driving scenarios. Recent advances have proposed modeling risk as a continuous field over space and time, offering richer representations that naturally encode spatial uncertainty, multi-agent interactions, and environmental context. However, systematic comparisons of such risk field formulations remain limited, and no unified theoretical foundation has emerged to guide their design.

This talk consists of two parts. In the first half, we present a comparative study of representative risk field models from the literature. Using real-world bird's-eye-view (BEV) datasets spanning diverse interaction scenarios including highway merging, occlusion, and stop-and-go traffic. We then evaluate these models through a unified set of field-level metrics assessing spatial coherence, temporal consistency, and behavioral alignment with human driving decisions. Our analysis reveals that different modeling assumptions yield distinct spatial risk structures, with significant implications for downstream planning and prediction tasks.

In the second half, we discuss emerging directions toward unified field modeling that integrates physics-informed principles with data-driven methods like PINNs. We briefly introduce how partial differential equation (PDE)-based formulations, widely used in macroscopic traffic flow theory, can be adapted for microscopic risk propagation in autonomous driving contexts. Preliminary results on specifically designed robustness benchmarks will be shared, along with open discussions in bridging traffic flow modeling with real-time motion planning for intelligent transportation systems.

Bios: Mr. Zian Wang Peter is a first-year MPhil student in the Department of Data and Systems Engineering at the University of Hong Kong, supervised by Prof. Chen Sun. He received the BEng degree in Electronic and Information Engineering from The Hong Kong Polytechnic University in 2025, where he was advised by Prof. Ivan Ho Wang-hei with experiences on embedded systems, vehicular technology, and robotics. His research now focuses on integration of data-driven methods with risk-aware frameworks for improved prediction and safe planning in autonomous driving, and interdisciplinary topics within intelligent transportation systems.